Detailed Lineage on Development of State and Territory Datasets Using the Multiple Lines

Detailed lineage on development of state and territory datasets using the Multiple Lines of Evidence (MLE) method

For each state or territory, intersection of the forest cover dataset supplied by the jurisdiction with other appropriate independent datasets identified:

-High confidence areas – areas where all the examined external or independent datasets agreed with the dataset recommended by the relevant state or territory agencies that the areas were forest or non-forest. No further assessment was required for these areas.

-Moderate confidence areas – areas where some of the examined external or independent datasets agreed with the dataset recommended by the relevant state or territory agencies that the areas were forest or non-forest. These areas needed further analysis through consideration of the reliability of the external or independent datasets. No further assessment was required for areas where highly reliable external or independent datasets agreed with the dataset recommended by the relevant state or territory. The remaining areas were identified as potential errors, and were either referred back to the relevant state or territory agency for validation, or validated in-house using various ancillary data including high-resolution imagery such as Bing Maps, Google Earth Maps, SPOT5 and LiDAR.

-Low confidence areas – areas where all the examined external or independent datasets disagreed with the dataset recommended by the relevant state or territory agencies that the areas were forest or non-forest. All such areas were identified as potential errors, and were either referred back to the relevant state or territory agency for validation, or validated in-house using various data including high-resolution imagery such as Bing Maps, Google Earth Maps, SPOT5 and LiDAR.

External or independent datasets used include:

1. State-wide Land and Tree Study (SLATS) – Two independent datasets were available, one covering NSW and the ACT and the other covering QLD. Foliage Projective Cover (FPC) values of 11 or greater (equivalent to crown cover 20% or greater) were considered as forest candidates. The National Vegetation Information System (NVIS) version 4.1 dataset was utilised to identify areas meeting the height definition required under NFI’s definition of forest in the SLATS datasets.
2. National Carbon Accounting System (NCAS) – this dataset uses a similar definition of forest to the one used by the NFI. However, the NCAS dataset also includes a minimum patch size (0.25ha) whereas the NFI does not.
3. Dynamic Land Cover Mapping (DLCM) – The following tree cover categories were considered as forest candidates in this dataset: Closed (70-100% crown cover), Open (30-70% crown cover) and Sparse (10-30% crown cover)

The MLE methodology is described in detail in a technical report “Improving Australia’s forest area estimate using a Multiple Lines of Evidence approach” downloadable here:

Table 1 and Table 2 below show possible combinations of the output when state or territory datasets are intersected with three or two external datasets. Areas of high confidence in the forest or non-forest allocation in state or territory datasets were not assessed further. Areas of moderate or low confidence in the forest or non-forest allocation in state or territory datasets were referred to the appropriate state or territory agencies for validation. Where state or territory agencies were unable to assist with the validation, in-house validation was carried out using high resolution imagery including Bing Maps, Google Earth Maps, SPOT5 and LiDAR.

Table 1: Example of full matrix combinations when a state or territory dataset is intersected with three other datasets

Table 2: Example of full matrix combinations when a state or territory dataset is intersected with twoother datasets

Validation assessment results were incorporated to give improved and high-confidence forest cover datasetsfor each state or territory.

Look-up tables translating the state or territory data to NFI forest types were used where provided. Where this information was not provided, it was derived by ABARES from translating Levels 5 and 6 of the National Vegetation Information System (NVIS) version 4.1 dataset into NFI forest types.

All input datasets were converted to 100m rasters of ESRI Grid format. In addition, the rasters were aligned with the reference raster in an Albers projection

Australian Capital Territory

The following four datasets were intersected as part of the MLE method:

1. Forests of Australia (2008) – recommended by the ACT Environment and Sustainability Development Directorate (ESDD) as the most up-to-date forest cover dataset for the territory. This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NSW SLATS (2010)
3. NCAS (2011)
4. DLCM (2010)

Table 1 above shows a full matrix of the output from intersecting 4 datasets, such as for the ACT. Areas of high confidence in forest or non-forest allocation in the territory’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the territory’s dataset were validated by ESDD and ABARES using various ancillary data, including high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT).

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in the ACT.

New South Wales

The following four datasets were intersected as part of the MLE method:

1. NVIS 4.1, CRAFTIand MVGs –NSW Office of Environment and Heritage (OEH) recommended deriving a forest cover dataset for NSW from interpreting Levels 5 and 6 of the NVIS 4.1 dataset. In northern parts of NSW, where the NVIS 4.1 dataset was deemed of poor quality, the Comprehensive Regional Forest Type Inventory (CRAFTI) dataset was recommended and used. In other parts of NSW where both the CRAFTI and NVIS 4.1 datasets were unavailable, the Major Vegetation Groups (MVG) dataset (supplied by the commonwealth Department of the Environment) was used. A composite of these three datasets was developed and formed the NSW forest cover dataset. This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NSW SLATS (2010)
3. NCAS (2011)
4. DLCM (2010).

Table 1 above shows a full matrix of the output from intersecting 4 datasets, such as for NSW. Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were referred to the NSW OEH agency for validation. ABARES assisted with the validation process using various ancillary data, including high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT).

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in NSW.

Victoria

The following three datasets were intersected as part of the MLE method:

1. Forests of Australia (2008) - recommended by Department of Environment and Primary Industries (DEPI) as the most up to date forest cover dataset for Victoria. This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NCAS (2011)
3. DLCM (2010)

Table 2 above shows a full matrix of the output from intersecting 3 datasets, such as for Victoria. Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were validated by ABARES using various data including NVIS 4.1, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT).

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in Victoria.

Tasmania

The following three datasets were intersected as part of the MLE method:

1. TASVEG 2.0 – recommended and provided by Department of Primary Industries, Parkes, Water and Environment (DPIPWE). ABARES used the look-up tables provided by DPIPWE to derive a forest and non-forest cover layer for the state.This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NCAS (2011)
3. DLCM (2010)

Table 2 above shows a full matrix of the output from intersecting 3 datasets, such as for Tasmania.Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were referred to DPIPWE for validation. DPIPWE validated a proportion of these areas and ABARES used various data including NVIS 4.1, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT) to validate the rest of the areas.

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in Tasmania.

South Australia

The following three datasets were intersected as part of the MLE method:

1. NVIS 4.1 – recommended by Department of Environment and Natural Resources (DENR) and Department of Primary Industries and Regions South Australia (PIRSA). ABARES used a look-up table developed in-house to translate Levels 5 and 6 of the NVIS 4.1 dataset into forest or non-forest.This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NCAS (2011)
3. DLCM (2010)

Table 2 above shows a full matrix of the output from intersecting 3 datasets, such as for South Australia. Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were referred to DENR and PIRSA for validation. ABARES assisted PIRSA with the validation, and used various data including Persistent Green – Vegetation Fractional Cover dataset developed by TERN, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT). Foliage Projective Cover (FPC) estimates (values 16 or greater) from the Persistent Green – Vegetation Fractional Cover dataset were used to discriminate forest cover in areas where the NVIS 4.1 dataset was of very coarse scale.

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in South Australia.

Western Australia

The following 3 datasets were intersected as part of the MLE method:

1. FMPA and NVIS 4.1 – The Department of Environment and Conservation (DEC) supplied a dataset covering the South West Forest Management Plan Area (SWFMPA) only and recommended using NVIS 4.1 for the rest of the state, based on ABARES look-up table translating Levels 5 and 6 of the NVIS dataset into forest or non-forest.This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NCAS (2011)
3. DLCM (2010)

Table 2 above shows a full matrix of the output from intersecting 3 datasets, such as for Western Australia. Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were validated by ABARES. ABARES used various data sources including Persistent Green – Vegetation Fractional Cover dataset developed by TERN, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR(IceSAT). Foliage Projective Cover (FPC) estimates from the Persistent Green – Vegetation Fractional Cover dataset were used to discriminate forest cover in areas where the NVIS 4.1dataset was of very coarse scale.

The validated dataset was then combined with theNational Plantation Inventory(NPI) dataset to capture plantation areas in Western Australia.

Northern Territory

The following 3 datasets were intersected as part of the MLE method:

1. NVIS 4.1 – Supplied and recommended by the Department of Natural Resources, Environment, the Arts and Sport (NRETAS), including a look-up table to distinguish forest from non-forest. This would be the dataset used for the NFI in the absence of application of the MLE method.
2. NCAS (2011)
3. DLCM (2010)

Table 2 above shows a full matrix of the output from intersecting 3 datasets, such as for the Northern Territory. Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were validated by NTRES and ABARES staff using various data sources including Persistent Green – Vegetation Fractional Cover dataset developed by TERN, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT). Individual NVIS 4.1 dataset polygons were assessed for forest cover and classed into: Forest, Split, Non-forest and Other Wooded Lands. The ‘Split’ category of polygons contained polygons with up to 60% forest by area, with the balance of their area being non-forest. Foliage Projective Cover (FPC) estimates (values 24 or greater) from the Persistent Green – Vegetation Fractional Cover dataset were used to discriminate forest and non-forest in the ‘Split’ category.

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in Northern Territory.

Queensland

The following 4 datasets were intersected as part of the MLE method:

1. NVIS 4.1 – recommended by Department of Science, Information Technology, Innovation and the Arts (DSITIA). ABARES used a look-up table developed in-house to translate levels 5 and 6 of the NVIS 4.1 dataset into forest or non-forest.
2. QLD SLATS (2010)
3. NCAS (2011)
4. DLCM (2010)

Table 1 above shows a full matrix of the output from intersecting 4 datasets, such as for Queensland.Areas of high confidence in forest or non-forest allocation in the state’s dataset were not assessed further. Areas of moderate or low confidence in forest or non-forest allocation in the state’s dataset were validated by DSITIA and ABARES staff using various data sources including QLD Regional Ecosystems, high-resolution imagery from Bing Maps, Google Earth Maps, SPOT5 and LiDAR (IceSAT). Individual NVIS 4.1 dataset polygons were assessed for forest cover and classed into: Forest, Split, Non-forest and Other Wooded Lands. The ‘Split’ category of polygons contains polygons with a mix of forest and non-forest, with each polygon containing up to 60% forest. FPC values from the SLATS dataset were used to discriminate forest from non-forest in the “Split’ polygons.

The validated dataset was then combined with the National Plantation Inventory (NPI) dataset to capture plantation areas in Queensland.

Forests of Australia (2013) – All the validated state and territory datasets described above were merged to create the Forests of Australia (2013) national dataset. The final step involved using the catchment-scale land-use dataset to identify and mask out (allocate as non-forest) areas in the national forest cover dataset that have the following land-uses: cropping, horticulture, irrigation, residential, industrial and utilities.