DEA Land Cover (Landsat)

Version:: 2.0.0
Type:: Derivative, Raster
Resolution:: 30 m
Coverage:: 1988 to Present
Data updates:: Yearly frequency, Ongoing

Overview

About

Digital Earth Australia (DEA) Land Cover translates over 35 years of satellite imagery into evidence of how Australia’s land, vegetation, and waterbodies have changed over time.

This version includes breaking changes

All tile grid references have been changed to refer to a new origin point. Learn more in the Version 2.0.0 changelog.

Access the data

For help accessing the data, see the Access tab.

See it on a map

DEA Maps

Explore data availability

DEA Explorer

Access the data on AWS

Data sources

Access the data on NCI Thredds

Data sources

View code examples

Code examples

Land Cover Explorer (Digital Atlas)

Web application

OGC web services

Web services

Esri web service: DEA Land Cover Level 3

Web services

Esri web service: DEA Land Cover Level 4

Web services

Key specifications

For more specifications, see the Specifications tab.

Technical name	Geoscience Australia Landsat Land Cover 30m
Bands	level3, level4
DOI	10.26186/149976
Currency	See currency and the latest and next update dates
Parent products	DEA Water Observations (Landsat), DEA Water Observations Statistics (Landsat), DEA Fractional Cover (Landsat), DEA Fractional Cover Percentiles (Landsat), DEA Tasseled Cap Percentiles (Landsat), DEA Mangroves (Landsat), DEA Intertidal Elevation (Landsat) v1.0.0, GA SRTM 1 second DEM, Landsat 5, 7, 8, and 9 NBART and Observational Attributes
Collection	Geoscience Australia Landsat Collection 3
Licence	Creative Commons Attribution 4.0 International Licence

Cite this product

Data citation

Tissott, B., Mueller, N., 2022. DEA Land Cover 25m. Geoscience Australia, Canberra. https://dx.doi.org/10.26186/146090

Paper citation

Lucas, R., Mueller, N., Siggins, A., Owers, C., Clewley, D., Bunting, P., Kooymans, C., Tissott, B., Lewis, B., Lymburner, L., Metternicht, G., 2019. Land Cover Mapping using Digital Earth Australia. Data. 4(4):143. https://doi.org/10.3390/data4040143

Publications

Lucas R, Mueller N, Siggins A, Owers C, Clewley D, Bunting P, Kooymans C, Tissott B, Lewis B, Lymburner L, Metternicht G. Land Cover Mapping using Digital Earth Australia. Data. 2019; 4(4):143. https://doi.org/10.3390/data4040143

Christopher J. Owers, Richard M. Lucas, Daniel Clewley, Carole Planque, Suvarna Punalekar, Belle Tissott, Sean M. T. Chua, Pete Bunting, Norman Mueller & Graciela Metternicht (2021) Living Earth: Implementing national standardised land cover classification systems for Earth Observation in support of sustainable development, Big Earth Data, 5:3, 368-390, DOI: 10.1080/20964471.2021.1948179

Description

Background

Land cover is the observed physical cover on the Earth’s surface including trees, shrubs, grasses, soils, exposed rocks, water bodies, plantations, crops, and built structures. An Australia-wide land cover product provides an understanding of how the different parts of the environment change and interrelate. Earth observation data recorded consistently over a period of time allows the observation of the state of land cover at a specific time and also the way that land cover changes by comparison between times.

Use constraints

DEA Land Cover is appropriate to use at the national scale where other more detailed information on land cover is not available. Where DEA Land Cover’s data conflicts with state or local datasets, those datasets should be considered authoritative.

What this product offers

DEA Land Cover provides annual land cover classifications for Australia using the Food and Agriculture Organisation’s (FAO) Land Cover Classification System (LCCS) taxonomy Version 2 (Di Gregorio and Jansen, 1998; 2005).

DEA Land Cover divides the landscape into six base land cover types, which are then further detailed through the addition of environmental descriptors (currently only accessible on DEA Maps). The structure is as follows.

Cultivated Terrestrial Vegetation
- Percentage of cover
- Life form (woody or herbaceous only)
Natural Terrestrial Vegetation
- Percentage of cover
- Life form (woody or herbaceous)
Natural Aquatic Vegetation
- Percentage of cover
- Life form (woody or herbaceous)
- Water seasonality
Artificial Surfaces
Natural Bare
- Percentage of bare cover
Water
- Water persistence
- Intertidal area

Applications

Annual Land Cover information can be used in a number of ways to support the monitoring and management of environments in Australia. These include, but are not limited to, the following areas of environmental monitoring, primary industries, and the interests and safety of the Australian community.

Environmental monitoring
- Ecosystem mapping
- Carbon dynamics
- Erosion management
Agriculture Sector
- Monitoring crop responses to water availability
- Understanding drought impact on vegetation
Community interests
- Map urban expansion within Australia
- Mapping impacts of natural disasters
- Bushfire recovery

Timeseries Perth urban expansion — a) Urban expansion of Quinns Rocks (Perth).

Timeseries Mt Beggary bushfire — b) Vegetation change and recovery following the Mt Beggary bushfire.

Timeseries inlet changes — c) Morphological changes of Shallow Inlet in Victoria.

Timeseries lake Amadeus — d) Annual variability of Lake Amadeus, a salt lake in the Northern Territory; the water persistence classification can be employed to track changes in its extent.

Figure 1. Examples of phenomena that can be tracked with the land cover product.

Technical information

DEA Land Cover is based on the globally applicable Food and Agriculture Organisation’s (FAO) Land Cover Classification System (LCCS) taxonomy Version 2 (Di Gregorio and Jansen, 1998; 2005). DEA Land Cover classifications have been generated by combining quantitative (continuous) or qualitative (thematic) environmental information (referred to as Essential Descriptors; EDs) derived from Landsat satellite sensor data. Several EDs have been generated previously by Geoscience Australia, including annual water summaries (Mueller et al., 2016), vegetation fractional cover (Scarth et al., 2010), mangrove extent (Lymburner et al., 2020) and the Intertidal Extent Model (ITEM; Sagar et al., 2017), whilst others have been developed more recently. These EDs have been combined to generate detailed, consistent, and expandable annual classifications of Australia’s land cover from 1986 through to the present.

DEA Land Cover consists of two datasets: The base (Level 3) classification, and the final (Level 4) classification which combines the base classes with their associated descriptors.

Detailed layer descriptions, including known issues, are provided in the following sections.

Level 3

Cultivated Terrestrial Vegetation (CTV)

Cultivated Terrestrial Vegetation (CTV) is associated with agricultural areas where active cultivation has been observed. In version 2.0, only herbaceous cultivation is shown and describes vegetation of strongly varying cover, ranging from bare (e.g. ploughed) areas to fully developed crops. Whilst the continental product describes land cover, interpretation is complicated as the same terminology is used to report on land use.

The definition of cultivated, and the difference from natural or semi-natural land covers, can be contentious, particularly as much of the Australian landscape is used for agricultural food production. This includes areas of Natural Terrestrial Vegetation (NTV) and Natural Aquatic Vegetation (NAV) that are grazed by stock and which can be regarded as either semi-natural or cultivated.

CTV in the DEA Land Cover map is associated with areas where management practices aimed at cultivation (including for grass production) are actively performed during the year being shown. These practices include crop planting and harvesting, fertilisation, and ploughing. These practices often lead to highly dynamic spectral signals within and between years but also regular transitions between vegetation of different cover amounts as well as bare soil. This also means that agricultural areas will transition between natural and cultivated covers as management practices transition an area between actively cropped or grazed, to areas left fallow, areas reduced to low cover due to climate effects such as drought, or to other covers depending on what the predominant conditions are through the year being shown.

Natural Terrestrial Vegetation (NTV)

Natural Terrestrial Vegetation (NTV) represents areas that have all or most of the characteristics of natural or semi-natural herbaceous or woody vegetation (based primarily on floristics, structure, function, and dynamics). These areas are identified as primarily vegetated, with either a photosynthetic vegetation fraction (PV) or non-photosynthetic fraction (NPV) greater than the bare soil fraction (BS) for at least two consecutive months. This approach considers that vegetation can exist in, and transition between PV and NPV states during the year. In effect, this approach classifies the landscape as primarily vegetated where the vegetated fraction of a pixel is greater than 30 %. Where the proportion of the landscape is less than 30 % vegetated, it is regarded as a sparsely vegetated or Natural Bare Surface, but the cover proportions can still be quantified. Urban areas that are vegetated (e.g. suburbs with trees) are associated with NTV if the pixel is at least 30 % vegetated but as Artificial Surfaces (AS) otherwise. The implementation allows areas of semi-natural vegetation (e.g. native grassland and pastureland) to be included in the NTV class.

Natural Aquatic Vegetation (NAV)

Natural Aquatic Vegetation (NAV) is associated primarily with wetlands that are dominated by woody or herbaceous vegetation (as with NTV). NAV is generally associated with swamps, fens, flooded forests, saltmarshes, or mangroves. Only mangroves are included in the current release.

Artificial Surfaces (AS)

Artificial Surfaces (AS) are areas of non-vegetated land cover created by human activities and are primarily represented by impervious surfaces (e.g. urban and industrial buildings, roads, and railways). These can be more readily identified when the area is larger than the spatial resolution (30 m) provided by the sensor. Open cut extraction sites are often included in AS. However, there is considerable misclassification of NS as AS in areas where vegetated cover is very low and very consistent through the year.

Natural Bare Surfaces (NS)

Natural Bare Surfaces (NS) are comprised primarily of unconsolidated (often pervious, e.g. mudflats and saltpans) or consolidated (e.g. bare rock or bare soil) materials. In Australia, the proportional area of Natural Bare Surfaces is relatively low and primarily confined to the deserts and semi-arid areas, river channels (e.g. dry riverbeds) and the coastline (e.g. mudflats and sand dunes). Much of the interior of Australia is sparsely vegetated and can be dominated by herbaceous (annual or perennial) or woody lifeforms.

Water

The Water class captures terrestrial and coastal open water such as dams, lakes, large rivers, and the coastal and near-shore zone.

Level 4

Classifications of Level 4

All categories from Level 3 and Level 4 algorithms are combined to give a single classification value for a given pixel.

The list below includes classes of different hierarchies, reflecting different levels of detail. The higher-level classes are used as fallback classes in cases where a confident detailed classification was not possible, and they are generally rare.

   Cultivated Terrestrial Vegetated

   Cultivated Terrestrial Vegetated: Woody

   Cultivated Terrestrial Vegetated: Herbaceous

   Cultivated Terrestrial Vegetated: Closed (> 65 %)

   Cultivated Terrestrial Vegetated: Open (40 to 65 %)

   Cultivated Terrestrial Vegetated: Open (15 to 40 %)

   Cultivated Terrestrial Vegetated: Sparse (4 to 15 %)

   Cultivated Terrestrial Vegetated: Scattered (1 to 4 %)

   Cultivated Terrestrial Vegetated: Woody Closed (> 65 %)

  Cultivated Terrestrial Vegetated: Woody Open (40 to 65 %)

  Cultivated Terrestrial Vegetated: Woody Open (15 to 40 %)

  Cultivated Terrestrial Vegetated: Woody Sparse (4 to 15 %)

  Cultivated Terrestrial Vegetated: Woody Scattered (1 to 4 %)

  Cultivated Terrestrial Vegetated: Herbaceous Closed (> 65 %)

  Cultivated Terrestrial Vegetated: Herbaceous Open (40 to 65 %)

  Cultivated Terrestrial Vegetated: Herbaceous Open (15 to 40 %)

  Cultivated Terrestrial Vegetated: Herbaceous Sparse (4 to 15 %)

  Cultivated Terrestrial Vegetated: Herbaceous Scattered (1 to 4 %)

  Natural Terrestrial Vegetated

  Natural Terrestrial Vegetated: Woody

  Natural Terrestrial Vegetated: Herbaceous

  Natural Terrestrial Vegetated: Closed (> 65 %)

  Natural Terrestrial Vegetated: Open (40 to 65 %)

  Natural Terrestrial Vegetated: Open (15 to 40 %)

  Natural Terrestrial Vegetated: Sparse (4 to 15 %)

  Natural Terrestrial Vegetated: Scattered (1 to 4 %)

  Natural Terrestrial Vegetated: Woody Closed (> 65 %)

  Natural Terrestrial Vegetated: Woody Open (40 to 65 %)

  Natural Terrestrial Vegetated: Woody Open (15 to 40 %)

  Natural Terrestrial Vegetated: Woody Sparse (4 to 15 %)

  Natural Terrestrial Vegetated: Woody Scattered (1 to 4 %)

  Natural Terrestrial Vegetated: Herbaceous Closed (> 65 %)

  Natural Terrestrial Vegetated: Herbaceous Open (40 to 65 %)

  Natural Terrestrial Vegetated: Herbaceous Open (15 to 40 %)

  Natural Terrestrial Vegetated: Herbaceous Sparse (4 to 15 %)

  Natural Terrestrial Vegetated: Herbaceous Scattered (1 to 4 %)

  Natural Aquatic Vegetated

  Natural Aquatic Vegetated: Woody

  Natural Aquatic Vegetated: Herbaceous

  Natural Aquatic Vegetated: Closed (> 65 %)

  Natural Aquatic Vegetated: Open (40 to 65 %)

  Natural Aquatic Vegetated: Open (15 to 40 %)

  Natural Aquatic Vegetated: Sparse (4 to 15 %)

  Natural Aquatic Vegetated: Scattered (1 to 4 %)

  Natural Aquatic Vegetated: Woody Closed (> 65 %)

  Natural Aquatic Vegetated: Woody Closed (> 65 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Woody Closed (> 65 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Woody Open (40 to 65 %)

  Natural Aquatic Vegetated: Woody Open (40 to 65 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Woody Open (40 to 65 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Woody Open (15 to 40 %)

  Natural Aquatic Vegetated: Woody Open (15 to 40 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Woody Open (15 to 40 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Woody Sparse (4 to 15 %)

  Natural Aquatic Vegetated: Woody Sparse (4 to 15 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Woody Sparse (4 to 15 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Woody Scattered (1 to 4 %)

  Natural Aquatic Vegetated: Woody Scattered (1 to 4 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Woody Scattered (1 to 4 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Herbaceous Closed (> 65 %)

  Natural Aquatic Vegetated: Herbaceous Closed (> 65 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Herbaceous Closed (> 65 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Herbaceous Open (40 to 65 %)

  Natural Aquatic Vegetated: Herbaceous Open (40 to 65 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Herbaceous Open (40 to 65 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Herbaceous Open (15 to 40 %)

  Natural Aquatic Vegetated: Herbaceous Open (15 to 40 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Herbaceous Open (15 to 40 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Herbaceous Sparse (4 to 15 %)

  Natural Aquatic Vegetated: Herbaceous Sparse (4 to 15 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Herbaceous Sparse (4 to 15 %) Water < 3 months (temporary or seasonal)

  Natural Aquatic Vegetated: Herbaceous Scattered (1 to 4 %)

  Natural Aquatic Vegetated: Herbaceous Scattered (1 to 4 %) Water > 3 months (semi-) permanent

  Natural Aquatic Vegetated: Herbaceous Scattered (1 to 4 %) Water < 3 months (temporary or seasonal)

  Artificial Surface

  Natural Surface

  Natural Surface: Sparsely vegetated

  Natural Surface: Very sparsely vegetated

  Natural Surface: Bare areas, unvegetated

  Water

  Water: (Water)

 Water: (Water) Tidal area

 Water: (Water) Perennial (> 9 months)

 Water: (Water) Non-perennial (7 to 9 months)

 Water: (Water) Non-perennial (4 to 6 months)

 Water: (Water) Non-perennial (1 to 3 months)

No data

Lifeform (NTV, NAV, and CTV; 2 classes)

Lifeform represents the dominant vegetation type of a primarily vegetated area, discriminating woody from non-woody (herbaceous) vegetation. The Woody Cover Fraction models woody as vegetation of at least 20 % canopy cover. Hence the dominant vegetation in areas designated as woody in this product is considered to be composed of shrubs and trees. However, where woody vegetation is not dominant in an area, the cover will be essentially herbaceous or bare. Hence some areas containing sparse trees or shrubs will likely be represented as herbaceous.

Vegetation Cover (NTV, NAV, and CTV; 5 classes)

Vegetation Cover is defined using the statistics of annual fractional cover of PV (for a calendar year). This relates to the uppermost foliage as observed from the Landsat satellite sensor, and describes the percentage of an area that is vegetated rather than bare.

Water Seasonality (NAV; 2 classes)

Water Seasonality refers to the typical hydrological conditions in NAV within a year and is relevant to both coastal and inland wetlands. The current implementation utilises the DEA Water Observations (WO) dataset, identifying hydro-periods for NAV areas where water is somewhat permanent (over 3 months) or temporary or seasonal (under 3 months).

Water State (Water; 1 class)

Water State establishes whether water is present in liquid form or as snow or ice. The current product only identifies areas where water is present as liquid for at least 20 % of observations (based on WOfS).

Water Persistence (Water; 4 classes)

Water Persistence (or hydro-period) describes the maximum duration (in months) that water is seen to be covering the surface in the year.

Intertidal (Water; 1 class)

Intertidal water refers to primarily non-vegetated aquatic areas with systematic tidal water variations.

Natural Surface (NS; 3 classes)

Natural Surface (i.e., bare gradation) describes the percentage of bare surface in areas which contain sporadic or little persistent green vegetation through the year. The percentage reflects that much of the remaining area is brown or dead vegetation. This is characteristic of the more arid parts of Australia.

Lineage

The FAO LCCS taxonomy (Figure 2) is hierarchical and consists of a dichotomous phase (Levels 1 to 3) and a modular phase (Level 4). In Level 1, vegetated and non-vegetated areas are first separated. These are then divided into terrestrial or aquatic categories to form Level 2. In the vegetated terrestrial category, cultivated and natural (including semi-natural) areas are differentiated. The non-vegetated category is further divided into Artificial Surface and Natural Surface. These non-vegetated Natural Surface include low vegetation cover and bare areas. This results in six base land cover categories (including the non-vegetated aquatic class from Level 2).

At Level 4, vegetated areas are further classified using information that differentiates lifeform (woody and herbaceous) and quantifies vegetation cover percent and water seasonality (for Natural Aquatic Vegetation). Natural Surface areas have information added (bare gradation) which describes the level of remaining vegetation present (sparse, very sparse, or not detectable). Non-vegetated aquatic areas (Water) are further described on the basis of their persistence (hydro-period) over a calendar year. The FAO LCCS differentiates water in different physical states (liquid or frozen; ice or snow); however, only liquid water is included in the current release.

Diagram showing the portion of the LCCS taxonomy which is implemented in DEA Land Cover.

Figure 2 (above). A diagram of the implementation of the FAO LCCS Version 2 classification within the DEA Land Cover product.

Diagram showing the data products which go into producing the Level 3 classification.

Figure 3 (above). Input data products used to produce the Level 3 classification.

Diagram showing input data products used to produce the Level 4 classification

Figure 4 (above). Input data products used to produce the Level 4 classification.

Software

References

Metternicht, G., Mueller, N., Lucas, R., Digital Earth for Sustainable Development Goals, Manual of Digital Earth, pp 443 - 471, Springer Singapore. https://doi.org/10.1007/978-981-32-9915-3_13

Quality

Limitations

DEA Land Cover is created by combining multiple layers that each describe features in the landscape. In doing so, the extents of each layer do not necessarily completely align, and some no-data points can cross between outputs. As a result, there are some Level 4 classes that only report detail to Level 3, as the details of cover fraction and water persistence do not have corresponding data in the respective datasets. This specifically relates the classes of Water, Natural Bare Surfaces, and Natural Aquatic Vegetation in areas near water bodies and the intertidal zone; however, the number of affected pixels is small.

Level 3 Class limitations

Cultivated Terrestrial Vegetation (CTV)

Managed plantations and some orchards and tree crops are not currently distinguishable from semi-natural or natural terrestrial vegetation and are not yet incorporated in the area of CTV. Reference can be made to Australia’s National Plantation Inventory. When savannah, semi-arid, and arid regions experience alternating browning and greening (as can occur due to fires, inundation, drought, and rainfall), this resembles the cyclical changes of cultivated land. This can cause areas of Natural Terrestrial Vegetation (NTV), Natural Bare Surfaces (NS), or Natural Aquatic Vegetation (NAV) to be misclassified as CTV. For example, the anomalously high levels of rainfall in 2010 led to some vegetation growth patterns that were classified as cultivated vegetation and these false positives reduced the precision of the class in that year. Several natural vegetation types, particularly in the monsoonal north, are misclassified as CTV due to burning, which can be associated with the indigenous management cycle. Saltmarshes, surface algae on mudflats, and aquatic vegetation on ephemeral inland floodplains can also be misclassified. Areas of bare soil exposed for long periods during the agricultural cycle or management activities and sparse vegetation (when areas are not in use or fallow such as during drought periods) can be misclassified as NS. This occurs particularly where areas have experienced low cover for prolonged periods within a year. The CTV machine learning algorithm has not been trained on CTV-woody areas, due to a lack of available training data, potentially leading to higher misclassification rates in these areas.

Natural Terrestrial Vegetation (NTV)

The NTV category can transition into the NS category between years when a highly variable Landsat satellite spectral signature is observed due to changes in vegetation productivity and moisture content. The transition of NTV to NS is particularly evident when comparing periods with rainfall above or below the average (e.g. 2010 was above; whereas, 2015 was below). Some confusion between surfaces that are primarily bare surfaces and non-photosynthetic vegetation remains and is also partly a function of the spectral reflectance of the underlying soil type. Vegetated areas shadowed by terrain can also be misclassified as non-vegetated. Areas of very low vegetation coverage are associated with the NS rather than NTV category, with the cover percentages shown as percentage bare. The assignment of vegetated suburbs to NTV is largely correct, but Artificial Surfaces (AS) located beneath the canopy are not currently represented in the product. Some confusion between CTV and NTV can occur because of natural seasonal changes in native vegetation. The edges of salt lakes may be misclassified as NTV.

Natural Aquatic Vegetation (NAV)

Currently only mangroves are mapped in the NAV class. Other vegetated natural landscapes (e.g. saltmarsh, river red gum forests in the Murray Valley, surface algae, and other inland wetlands) where water significantly influences edaphic conditions of the substrate are not mapped. This is because in the current implementation the water mask is included to assist in differentiating vegetation and non-vegetation, as the presence of water creates excess noise in the underlying Fractional Cover product. To reduce this noise, the DEA Water Observations product is used as a water mask in the Fractional Cover product, and hence it is unlikely to produce the combination of vegetation and water required for the NAV class.

Artificial Surfaces (AS)

The Artificial Surfaces machine learning algorithm works best in dense urban areas. Misclassification occurs with Natural Bare Surfaces (NS), particularly in arid and semi-arid regions, open cut mine sites, salt lakes and pans, sand dunes, and beaches. This is attributed to similarities in the variance of spectral signatures over a year. The Australian Bureau of Statistics - Urban Centre and Locality mask is applied to reduce some of these misclassifications in rural areas. Urban areas with high concentrations of surrounding vegetation may not be correctly classified as AS. Misclassification occurs in some cultivated areas due to the predominance of sparse vegetation, or when land is left fallow for most of the year. Cloud and data quality issues can lead to the incorrect assignment of other land cover classes to AS, such as in south-west Tasmania. Within dense urban centres with high-rise buildings such as the central business districts of Sydney or Melbourne, some areas are misclassified as Water due to the shadows of the high rise buildings. In some industrial areas, buildings can be misclassified as no data due to the bright surface reflectance having a similar signature to cloud.

Natural Bare Surfaces (NS)

Land used for agriculture may be associated with the NS class if ploughing or tillage has occurred and the vegetation cover remains low during the calendar year. Areas mapped as NS may temporarily transition to or from CTV under certain circumstances (e.g. during drought). This occurs because of the predominance of temporary vegetation (e.g. short crop life spans), desiccation, removal during dry periods, or management practices (e.g. land left as fallow). Where acquisition of cloud-free Landsat satellite sensor data has been infrequent, less information on the inter-annual variability of vegetation is provided. Where cloud cover is extensive (e.g. in Tasmania), the number of scenes for land cover classification is reduced and this can compromise classification of NS, particularly where bare surfaces alternate with vegetation. Confusion between natural surfaces and urban areas occurs because of similar spectral variation over an annual period.

Water

Areas of artificial and natural water are not differentiated, although the extent of artificial water is mapped within the existing Bureau of Meteorology Geofabric product. Due to the 30 m pixel size, rivers and water courses that cover less than a pixel, or with highly vegetated riverbanks are not captured. This results in a patch-like representation of narrower rivers. Areas of dark, wet soil are often misclassified as water, including in cultivated areas and mud flats. Some pixels surrounding water bodies have no classification, due to having no valid data from the Fractional Cover product, and the Water Observations product being unable to determine if the pixels are wet or dry. Classification of pixels over the ocean should not be used from Land Cover — the water persistence is often incorrect, and the land cover classification should be applied to land-only pixels. This Land Cover product should be used with an ocean mask for coastal tiles.

Level 4 Class limitations

Lifeform

Woody discrimination is implemented using the Woody Cover Fraction product (Liao et al, 2020) which models woody cover from inputs of LiDAR including ICESat/GLAS, L-band SAR, field observation, and Landsat satellite data. Issues arise in this dataset in areas dominated by short, woody vegetation such as heathland, and swampy regions with underlying water. Areas of woody savannah are also under-represented due to the influence of the herbaceous underlying cover dominating the observation.

Vegetation Cover

The cover of vegetation is derived from the Fractional Cover product (Scarth et al, 2010), and as such it reflects the limitations of that product, mainly difficulty with measurement of non-photosynthetic vegetation, and noise due to the presence of water in a pixel. Therefore, arid areas can be difficult to fully analyse for cover, leading to misclassifications between NTV, NS, and CTV where cover is sparse (lower than 15 %).

Water Seasonality

This product does not yet identify consecutive months but rather the frequency of wet observations in the year, based on the Water Observations product. Therefore, monthly statements are unlikely to be consistent across the continent. Mangroves are currently the only consistently identified NAV and water cannot be easily observed beneath their dense canopies. Hence, the hydro-period (and seasonality metrics) should be treated with some caution.

Water State

Mapping of ice and snow states is yet to be undertaken. Hence the Water State class is currently limited to liquid only.

Intertidal

The intertidal areas carry the limitations of the DEA Intertidal Extents (Landsat) product:

Intertidal areas are determined as a coastal mask, which currently identifies some non-tidal inland water bodies as tidal because of intra-annual changes in water extent.
This product is static and cannot be used to demonstrate change between two annual continental products.

Bare Gradation

The Bare Gradation class is produced from the Fractional Cover product. Unlike the Vegetation Cover class, Bare Gradation is calculated from the median bare fraction, rather than consecutive monthly green and non-photosynthetic fractions. As such the bare fraction can be as low as 20%, however this does not imply that the remaining fraction is healthy vegetation. Rather, the remaining fraction is a mix of brown, dead, and green vegetation, with intermittent green periods through the year, reflective of arid area vegetation types.

Earth Observation limitations

To generate the land cover classification for each calendar year, annual (January to December) statistics derived from Landsat-5, -7, -8, and -9 observations are currently used, with each satellite sensor potentially observing the Australian landscape every 16 days. This brings an intrinsic limitation to land cover mapping as persistent cloud in some regions reduces the number of useable observations. This is particularly evident in Tasmania, and northern Australia during the monsoon period, where areas may not be observed for extended periods and parts or all of the intra-annual land cover cycle may be missed due to persistent cloud. These limitations can lead to misclassifications of land cover, particularly in dynamic environments. In a future release, a confidence layer will be included to help identify areas with poor observation frequency or other factors impacting the classification.

An additional limitation of the Landsat series is the availability of data due to the ageing of each satellite. Landsat 5 was operational for over 25 years, but for much of the later years, data were only acquired when sunlight directly illuminated its solar panels. This limited its operation across Australia, with coverage being seasonally dependent, and contracting north to a minimum in winter. In its last years, the winter coverage usually only covered the northern coasts of Australia. Landsat 5 ceased regular operations over Australia in 2011, leaving just Landsat 7 operational until Landsat 8 was launched in 2013. Landsat 7 began service in 1999 as a replacement for Landsat 5. Initially Landsat 5 was switched off, but when Landsat 7 suffered a serious problem in 2003 due to the failure of its scan-line corrector (termed SLC-Off), Landsat 5 resumed service. The SLC-Off failure of Landsat 7 resulted in severe striping across every image from mid-2003 onwards, and this striping is apparent in subsequent derived products. Landsat 8 has operated well since its launch in 2013. Landsat 9 has operated since 2022 with improved sensitivity, noise characteristics, and data correction in comparison to the earlier sensors.

The result of the availability of these satellites is that the most consistent data availability occurs when two satellites are in operation (which occurred in most of the 2003 to present period). The least data availability is in 2011 to 2012 when only Landsat 7 was available, and its data quality was impaired by the SLC-Off striping issue. The overall data availability for the Landsat satellites is shown in Table 1. The datasets used in this analysis are shown in Table 2.

Table 1 (below). Overall Landsat sensor data availability in the DEA data cube used for the Land Cover classification.
Years	Satellite and sensor	Issues	Impact on LCCS product
1986	Landsat 5 Thematic Mapper (TM)	Only part of the year available in DEA.	Very little data available. Year is omitted.
1987 to 1999	Landsat 5 TM	Normal operation	Only single source of data is available.
1999 to early 2003	Landsat 5 TM	Switched off with availability of Landsat 7.	Only single source of data is available.
1999 to early 2003	Landsat 7 Enhanced Thematic Mapper (ETM+)	Normal operation.	Only single source of data is available.
Late 2003 to early 2011	Landsat 5 TM	Gradually degrading operation and increasing observation failure. Total failure in early 2011.	Landsat 5 with Landsat 7 to fill in missed areas (clouds).
Late 2003 to present	Landsat 7 ETM+	Operating with SLC-Off stripes in imagery.	Fill in for missed areas in Landsat 5 and 8. Only data available from early 2011 to early 2013.
Early 2013 to present	Landsat 8 Optical Land Imager (OLI)	Normal operation.	Landsat 8 with Landsat 7 to fill in missed areas (clouds).

Table 2 (below). Datasets within DEA used to provide Essential Descriptor information for the DEA Land Cover classification.
Product	Used for	Dataset reference
Fractional Cover (FC)	Vegetation	Scarth et al. (2010), Schmidt et al. (2010)
Water Observations from Space (WOfS)	Aquatic	Mueller et al. (2016)
Intertidal Extent Model (ITEM)	Aquatic	Sagar et al. (2017)
National Mangrove extent	Aquatic	Lymburner et al. (2019)
Machine learning applied to annual geometric medians (geomedians) and Median Absolute Deviations (MADs)	Cultivated	Roberts et al. (2017) (geomedians and MADs); Chua (cultivated) DEA (in draft)
Machine learning applied to geomedians, MADs, and Woody Cover Fraction	Artificial surfaces	Tan and Chua, DEA (in draft)
Woody Cover Fraction (WCF)	Lifeform	Liao et al. (2020)

Inconsistent data presence in three Western Australia tiles

An issue has been identified with data presence in three tiles in Western Australia, particularly between 1988 and 2005. The affected tiles are x39y49, x40y47, and x41y45. To visualise these locations, see the DEA Explorer (ga_ls_landcover_class_cyear_3).

This issue arises due to the Ground Quality Assured (GQA) value falling below the acceptable threshold set by the DEA for products derived from Landsat. The root cause of this anomalous behaviour lies in the Analysis Ready Data (ARD), which serves as the source for this Land Cover product. The areas affected by this issue correspond to regions where ground control points may experience movement over time, such as shifting sand dunes, leading to inconsistencies in the orthorectification process.

timeseries missing data tiles

Figure 1. Time series animation showing the frequency of missing data in the tiles x39y49, x40y47, and x41y45. On the left is a true colours time series of the yearly geomedian. On the right is a time series of the yearly clear observation count.

Accuracy

A validation assessment has been undertaken for both of the versions of Land Cover: the former Collection 2 (C2; i.e. Version 1), and the current Collection 3 (C3; i.e. Version 2). The below section outlines the accuracy of both versions to assist users in understanding the differences between them. The validation metrics reported were produced for Level 3 and they integrate the results from the validation of the sub-components of Level 1, the Artificial Surface model, and the Cultivated Terrestrial Vegetation model.

Summary of differences between Land Cover C2 and C3

Collection 3 generally aligns with Collection 2 in terms of classification performance, with trends across time periods and classes being mostly consistent. However, some deviations from this pattern have been observed across different locations and times, particularly in the CTV class.
Collection 3 classification shows better overall consistency across different validation methods.
An improvement in AS classification was observed in C3. More urban areas appear to be correctly identified, although there is a slight increase in false positive identification of urban areas in the central Australian desert and other sandy regions.
Slight improvement is seen in Woody Cover for the NTV classification.
There is a substantial increase in Landsat 7 stripe artefacts in C3 due to Landsat 8 scenes being filtered out due to poor geometric quality assessments.
There is a general increase in the number pixels with no data surrounding water bodies in C3.
Misclassification of water persistence over the ocean is observed in C3.

Improvement artificial class Degradation water persistence, Level 4 — a) Improved Artificial Surface (AS) classification in the greater Melbourne area can be seen in Collection 3; stripe artifacts and incorrect water persistence classification can also be observed.

Improvement woody cover, Level 4 — b) Pine plantation near Kinglake West, Victoria. In Collection 3 (C3; version 2) the plantation areas are classified as being > 65% Woody Cover, more closely reflecting the land cover of the area compared to Collection 2.

Degradation cultivated, Level 4 — c) Agricultural region in the Cassowary Coast, northern Queensland, where degradation in CTV classification in Collection 3 can be seen. This degradation is primarily due to misclassification of cultivated land as herbaceous Natural Terrestrial Vegetation (NTV).

legend Level 4

Figure 2 (above). Animations showing examples of differences in Level 4 classification between Land Cover C3 and C2.

Degradation false positives artificial, Level 3 — a)The town of Portland, Victoria, located in the eastern part of the captured region, demonstrates improved Artificial Surface (AS) classification of urban areas in Collection 3. However, some sandy terrains to the southwest of the town are incorrectly classified as AS rather than Natural Bare Surface (NS).

Improvement, cultivated better classified, Level 3 — b) Classification differences between Collection 2 and Collection 3 are evident across agricultural regions in southern Western Australia. Improved classification of agricultural areas is observed in the north, while degraded classification, characterized by noisier and less defined areas, is seen in the east.

Degradation missing pixels around waterbodies, Level 3 — c) Lake Eildon, Victoria, is an example of a lake affected by the increased number of pixels with missing values found at the edges of water bodies in Collection 3.

legend Level 3

Figure 3 (above). Animations showing examples of differences in Level 3 classification between Land Cover C3 and C2.

Collection 2 validation

The product was validated using 6,000 points spatially distributed across Australia for the years 2010 and 2015, resulting in a total of 12,000 samples. These points were created using a stratified random sampling approach that was adjusted for oversampling. After removing points with ‘No Data’ and spurious values, the total number of valid points was 11,750. The sample points were divided into clusters for visual assessment against the classification outputs and individually assessed by a pool of 10 people. To compare individual biases among the assessors, an additional set of validation points was created and evaluated by all assessors. Table 3 contains the overall accuracy for all classes. The term ‘support’ refers to the number of validation points used in the calculation of that accuracy value.

Table 3 (below). Overall accuracy of DEA Land Cover.

Overall accuracy of DEA Land Cover is 80%. 2010 accuracy is 82% and 2015 accuracy is 78%.

The results, shown in Table 4, also provides per-class accuracy information. ‘Precision’ refers to the ability of a classification model to return only relevant instances, while ‘Recall’ refers to the ability to identify all relevant instances. The ‘F1 score’ combines precision and recall to provide an overall measure of accuracy. Classes such as Artificial Surfaces (AS), Natural Aquatic Vegetation (NAV), and Water exhibited high accuracies. However, classifying Cultivated Terrestrial Vegetation (CTV) and Natural Bare Surfaces (NS) proved challenging, resulting in the lowest accuracies among the six classes presented.

Table 4 (below). Per-class accuracy information for DEA Land Cover.

Table showing accuracy per class, including precision, recall, F1 score, and support values per class.

Table 5 indicates that where assessors could identify a predominant land cover (i.e., not ‘mixed’ pixels or ‘unsure’), there was 75% agreement among all assessors.

Table 5 (below). Inter-assessor agreement.

Table showing the agreement between assessors.

Collection 3 validation

To understand the extent of change between the versions, validation against three data sources was undertaken. These were the validation points reused from Collection 2, with the addition of point attributes from Köppen Climate Zone and State/Territory to facilitate segment analysis; the GLANCE (Global Land Cover Estimation) global dataset of independent “ground truth” data; and the Land Cover Collection 2 data.

To account for the additional validation points, the Collection 2 (C2) validation was rerun and compared with the results from Collection 3 (C3).

Both C3 and C2 showed overall degradation in 2010 compared to 2015, a trend propagated from the Level 1 and individual ML models results. The performance metrics with the validation points show an overall alignment of the two collections, with minimal differences across categories, except for the Recall values of Cultivated Terrestrial Vegetation (CTV) and Artificial Surface (AS), which are both higher in C3 for the year 2010 (see Tables 6 and 9).

C3 demonstrated a more reasonable classification on the GLANCE dataset, particularly for the Natural Terrestrial Vegetation (NTV) class, which can be intuitively interpreted as the Agriculture category of the GLANCE dataset (see Tables 7, 8, 10, and 11).

C3 exhibited better consistency in results against the validation points and the GLANCE data. However, some misalignment between ground truth datasets is present in both collections for certain classes (e.g. Natural Bare Surfaces; NS). Some classes, such as Artificial Surface (AS), have too few validation points to be considered statistically significant. Hence, the classification metrics should be interpreted with caution and understood in relative terms, as a way to highlight the differences between C2 and C3, rather than the absolute performance of each.

When C3 was compared directly with the output of C2 (Table 16), it showed good agreement for most of the Level 3 classes. Two notable exceptions were the Cultivated Terrestrial Vegetation (CTV) class and the Artificial Surface (AS) class. For CTV, the performance against the validation data did not suggest a substantial difference in overall accuracy between C2 and C3; thus, the misalignment observed in Table 16 could be due to possible inconsistencies in classification between the two collections, but not necessarily an improvement or degradation. The difference observed in the AS classification may be influenced by the increased accuracy of C3.

In the classification metrics tables below, two types of averages are used to summarise the performance across different classes: Macro-Average and Weighted-Average. The Macro-Average is the unweighted mean of each metric, calculated independently for all classes. Each class contributes equally, regardless of its frequency in the dataset. This average should be interpreted as ‘unbiased’ due to the highly skewed nature of the validation points. The Weighted-Average, on the other hand, is the average of each metric weighted by the support (\(TP + FN\); i.e., True Positives and False Negatives) for each class. The weight reflects the proportion of each class in the ‘Truth’, meaning that more frequent classes have a greater impact on the overall average.

Specifications ✨

Product ID

The Product ID is ga_ls_landcover_class_cyear_3. This ID is used to load data from the Open Data Cube (ODC), for example dc.load(product="ga_ls_landcover_class_cyear_3", ...)

Bands

Bands are distinct layers of data within a product that can be loaded using the Open Data Cube (on the DEA Sandbox or NCI) or DEA’s STAC API.

	Type	Units	Resolution	No-data	Aliases	Description
level3	uint8	-	30 m	255	-	The base Level 3 Land Cover classification.
level4	uint8	-	30 m	255	full_classification	All Level 3 and Level 4 classes are combined to give a single classification value.

For more information on these bands, see the Description tab.

Product information

This metadata provides general information about the product.

Product ID	ga_ls_landcover_class_cyear_3	Used to load data from the Open Data Cube.
Short name	DEA Land Cover (Landsat)	The name that is commonly used to refer to the product.
Technical name	Geoscience Australia Landsat Land Cover 30m	The full technical name that refers to the product and its specific provider, sensors, and collection.
Version	2.0.0	The version number of the product. See the History tab.
Lineage type	Derivative	Derivative products are derived from other products.
Spatial type	Raster	Raster data consists of a grid of pixels.
Spatial resolution	30 m	The size of the pixels in the raster.
Temporal coverage	1988 to Present	The time span for which data is available.
Update frequency	Yearly	The expected frequency of data updates. Also called ‘Temporal resolution’.
Update activity	Ongoing	The activity status of data updates.
Currency	See the Currency Report	Currency is a measure based on data publishing and update frequency.
Latest and next update dates	See the Currency Report	See Table B of the report.
DOI	10.26186/149976	The Digital Object Identifier.
Catalogue ID	149976	The Data and Publications catalogue (eCat) ID.
Licence	Creative Commons Attribution 4.0 International Licence	See the Credits tab.

Product categorisation

This metadata describes how the product relates to other products.

Parent products	DEA Water Observations (Landsat), DEA Water Observations Statistics (Landsat), DEA Fractional Cover (Landsat), DEA Fractional Cover Percentiles (Landsat), DEA Tasseled Cap Percentiles (Landsat), DEA Mangroves (Landsat), DEA Intertidal Elevation (Landsat) v1.0.0, GA SRTM 1 second DEM, Landsat 5, 7, 8, and 9 NBART and Observational Attributes
Collection	Geoscience Australia Landsat Collection 3
Tags	geoscience_australia_landsat_collection_3, land, vegetation, water, urban

Access

Access the data

DEA Maps	See it on a map	Learn how to use DEA Maps.
DEA Explorer	Explore data availability	Learn how to use the DEA Explorer.
Data sources	Access the data on AWS Access the data on NCI Thredds	Learn how to access the data via AWS.
Code examples	View code examples	Learn how to use the DEA Sandbox.
Web application	Land Cover Explorer (Digital Atlas)	Learn how to use the Land Cover Explorer.
Web services	OGC web services	Learn how to use DEA’s web services.
Web services	Esri web service: DEA Land Cover Level 3	Learn how to create a map using Digital Atlas.
Web services	Esri web service: DEA Land Cover Level 4	Learn how to create a map using Digital Atlas.

History

Version history

Versions are numbered using the Semantic Versioning scheme (Major.Minor.Patch). Note that this list may include name changes and predecessor products.

v2.0.0	-	Current version
v1.0.0	of	DEA Land Cover (Landsat)

Changelog

Version 2.0.0

This latest version introduces improvements to the DEA Land Cover product, enhancing the translation of over three decades of satellite imagery into evidence of how Australia’s land, vegetation and waterbodies have changed over time.

Breaking change: Shift in grid origin point — The south-west origin point of the DEA Summary Product Grid has been shifted 18 tiles west and 15 tiles south. Therefore, all tile grid references have been changed. For instance, a tile reference of x10y10 has changed to x28y25.
Breaking change: Data structure changes — Land Cover version 1.0 contained several TIF files including waterstt-wat-cat-l4a.tif, watersea-veg-cat-l4a-au.tif, waterper-wat-cat-l4d-au.tif, lifeform-veg-cat-l4a.tif, inttidal-wat-cat-l4a.tif, canopyco-veg-cat-l4d.tif, and baregrad-phy-cat-l4d-au.tif. These will not be provided in Land Cover 2.0 because the classifications are instead contained within the level4.tif file, allowing users to filter the results as required. An RGB TIF colour representation of the Land Cover classifications is also no longer available. Instead, a style file will be provided that can be uploaded to your GIS software which will apply the same colour representations of the classifications in the Level 3 and Level 4 TIF files.
Improved cloud masking to reduce noise — The masking of clouds and cloud shadows applied to DEA Land Cover model’s input data was improved, resulting in fewer noisy pixels in the final Land Cover product.
Landsat 9 included — Landsat 9 is included from 2022 onwards, leading to improved performance compared to the standalone Landsat 8 product. This is due to the increased number of observations from the combined Landsat 8 and 9 sensors.
Machine Learning Upgrade — The settings for machine learning (ML) methods and algorithms have been changed to improve performance. This is for woody cover, cultivated vegetation, and artificial surface classification. The Urban model uses a different ML approach, shifting from a pixel-based to a raster-based method, and a new model was trained using collection 3 data. The cultivated model maintains the same approach, but the input features were re-engineered using Geoscience Australia’s collection 3 data, and the model was re-trained. The woody cover model also follows the same approach, with the model being re-trained using Geoscience Australia’s collection 3 data.
Pixel resolution — The pixel resolution for Land Cover 2.0 has changed from 25m to 30m resolution.
Product ID Update — The Land Cover product ID has been updated to reflect the new ARD collection being used. It was changed to ga_ls_landcover_class_cyear_3. This ID is used to load data from the Open Data Cube (ODC).
‘No data’ value changed — The ‘no data’ value has been changed from ‘0’ to ‘255’ to be consistent with other DEA derivative products.
Level 4 classification update — The level 4 classifications have been updated to include the class ‘Cultivated Terrestrial Vegetated: Woody’.
Collection Upgrade — Land Cover version 2.0 is based on Geoscience Australia’s latest Landsat collection 3 analysis ready data.

FAQs

Frequently Asked Questions

Credits

Acknowledgments

DEA acknowledges the contribution of several collaborators who helped to create this Land Cover product.

CSIRO — Anna Richards, Becky Schmidt, Kristen Williams
UNSW — Graciela Metternicht
Department of Agriculture, Water and Environment — Jane Stewart, Lucy Randall, Lucy Gramenz
ANU — Albert van Dijk
Tasmania DPIPWE — Lindsay Mitchell

License and copyright

Released under Creative Commons Attribution 4.0 International Licence.