- About Land Cover
- What is the Dynamic Land Cover Dataset?
- Development of the Dynamic Land Cover Dataset
- Accuracy of the Land Cover Map
- Interpreting the trend in annual Enhanced Vegetation Index (EVI) values
- Caveats for interpreting the EVI trends
- How to interpret the trend in the annual maximum EVI values
- How to interpret the trend in the annual minimum EVI values
- How to interpret the trend in the annual mean EVI values
- Common Names
- Glossary of ISO terms
Land cover is the observed biophysical cover on the Earth’s surface including trees, shrubs, grasses, soils, exposed rocks and water bodies, as well as anthropogenic elements such as plantations, crops and built environments. Land cover changes for many reasons, including seasonal weather, severe weather events such as cyclones, floods and fires, and human activities such as mining, agriculture and urbanisation.
Remote sensing data recorded over a period of time allows the observation of land cover dynamics. Classifying these responses provides a robust and repeatable way of characterising land cover types. A key aspect of land cover is vegetation greenness. The greenness of vegetation is directly related to the amount of photosynthesis occurring, and can be measured with an index such as the Enhanced Vegetation Index (EVI) hence different vegetated land cover types can be distinguished.
The Dynamic Land Cover Dataset is the first nationally consistent and thematically comprehensive land cover reference for Australia. It provides a base-line for reporting on change and trends in vegetation cover and extent. Information about land cover dynamics is essential to understanding and addressing a range of national challenges such as drought, salinity, water availability and ecosystem health.
The new Dataset presents a synopsis of land cover information for every 250m by 250m area of the country from April 2000 to April 2008. The classification scheme used to describe land cover categories in the Dataset conforms to the 2007 International Standards Organisation (ISO) land cover standard (19144-2). The Dataset shows Australian land covers clustered into 34 ISO classes. These reflect the structural character of vegetation, ranging from cultivated and managed land covers (crops and pastures) to natural land covers such as closed forest and open grasslands.
In addition to the land cover map, the Dynamic Land Cover Dataset contains 3 layers that show the trend in annual greenness characteristics between 2000 and 2008. These trend layers highlight areas that are becoming greener or less green over time.
The source data for the Dataset is a time series of EVI data from the Moderate Resolution Imaging Spectroradiometer (MODIS) on the Terra and Aqua satellites operated by NASA. The time series includes 186 snapshots of vegetation greenness for each 250m by 250m area across the continent over an 8 year period. An example of the time series displayed by different land cover types is shown in Figure 1.
The EVI time series for each 250m by 250m area was characterised using 12 time series coefficients which describe the statistical, phenological and seasonal characteristics of the land cover. A clustering approach was applied to the 12 coefficients to define homogenous regions with similar greenness dynamics over time.
Regions that showed similar greenness characteristics over time were labelled using information derived from the 2009 Catchment Scale Land Use of Australia dataset and Native Vegetation Information System dataset provided by ABARES. The labelling process is described in detail in the technical report.
State agencies provided more than 25000 field validation sites to assess the accuracy of the Dataset. As land cover classes are not generally clear-cut, but merge gradually from one to the other, a fuzzy-logic system (Zhang and Foody, 1998) was used to compare the 34 Dynamic Land Cover Dataset classes with the field data on a sliding scale. The classes of the sliding scale are:
- exact match such as trees open versus trees open,
- very similar such as trees open versus trees scattered,
- moderately similar such as trees open versus shrubs open,
- somewhat similar such as trees open versus shrubs closed, and
- complete mismatch such as trees open versus irrigated graminoids.
The match between the field data and the Dataset was exact in 30 per cent of cases, very similar in 35 per cent of cases, moderately similar in 10 per cent of cases, somewhat similar in 18 per cent of cases and a complete mismatch in 7 per cent of cases. These results show a high degree of consistency between the Land Cover Map and extensive independent field-based datasets. The accuracy of the land cover map across the main land cover types is shown Figure 2 and and the accuracy of the land cover map across all 34 ISO classes is shown Figure 3.
Each of the three trends represents a different aspect of the photosynthetic behaviour of the vegetation within that pixel over a year. The trend data was calculated as the slope of the line for a linear regression of the annual minimum EVI values (lowest 3 EVI values for each year) (Figure 4), annual maximum EVI values (highest 3 EVI values for each year) (Figure 5) and annual mean EVI values (average of all EVI values for that year) (Figure 6).
The trend data can be used to identify areas where vegetation is showing inter-annual changes over the 8 year period of observation. The factors that may be driving that change are listed in Table 3. In some instances the trend data will identify vegetation that has been impacted by specific events such as bushfires and tropical cyclones. In other instances the trend data will identify vegetation response to more gradual processes such as drought or changes in ground water availability. It is important to note that the trend data play a valuable role in detecting change, however additional data is required to identify which driver(s) is leading to the changes in greenness. In instances such as severe tropical cyclones and major bushfires, identifying the drivers of change is straightforward, however there are many instances where the trend data is detecting changes at the large paddock, regional and national scale where additional information is needed to characterise the processes that are driving the change.
|Broad scale driver||Specific Driver||Human system, Natural system|
|Plant available water|
|Timing of rainfall||Natural|
|Amount of rainfall||Natural|
|Soil organic C content||Both|
|Soil N,P,K content||Both|
|Removal of woody vegetation|
¹ This includes both the use of hyporheic groundwater by riparian vegetation and the use of phreatic groundwater by other ground water dependent vegetation
² The vegetation type/community that exists at any location is an emergent property of some combination of the other drivers.
Caveat 1. Extreme caution should be applied in linking any value judgement to the trend data. If there has been a large green flush on a central Australian floodplain river near the start of the time series and all three trends have been in decline since then, it doesn’t necessarily mean that the river is in ‘bad’ condition, it may simply mean that there hasn’t been a significant flood event since then. Likewise, an increase in greenness may represent an increase in the presence of invasive weeds rather than any improvement in productivity or condition.
Caveat 2. Trend data are easily skewed by extreme values, which is particularly problematic for the minimum and maximum trends that are by definition based on extreme values. This means that if an event that causes a major increase (for example a flood pulse on central Australian floodplain rivers) or major decrease (a severe bushfire) occurs at the start or end of the time series it will have a strong influence on the trend in the maximum or minimum. Trend data are of limited use when analysing episodic rather than seasonal systems.
Caveat 3. Interpretation of greenness dynamics needs to be done within the conceptual framework that acknowledges all factors that may be influencing greenness in that vegetation community and climatic/landscape setting. This means that the trend data is a very useful tool for flagging areas that are trending upwards or downwards more strongly than similar vegetation types in the same climate and landscape setting. Once these areas are flagged more detailed analysis is required to identify which factors (anthropogenic or otherwise) are driving the observed change.
Caveat 4. In areas where pixels contain little or no vegetative cover i.e. salt lakes and permanent water bodies, the EVI is very noisy and all trend data are meaningless.
Caveat 5. The trend data for an individual pixel has limited statistical significance, however if many pixels of the same class show the same trend, particularly if they are colocated, the statistical significance increases rapidly.
The annual maximum EVI value is related to the maximum leaf area1 (including both canopy cover and ground cover) observed at any point during a year. Irrigated summer crops may reach EVI values in excess of 8000, whereas the maximum EVI over the Nullarbor plain is around 2000. The trend in the maximum EVI value is very sensitive to individual ‘good’ years, so it can be misleading when applied to systems that are episodic rather than seasonal in nature. The data can potentially be used to assess whether maximum leaf area for a specific area is increasing or decreasing. This is a qualitative rather than quantitative measure i.e the maximum EVI is observed to drop from year to year, but the necessary ancillary data may not be available to link that to a absolute value for change in leaf area index.
In areas that are subject to more than 16 consecutive days inundation, the trend in the annual minimum may be skewed by flood events. Aside from these areas, annual minimum EVI relates to the minimum observed levels of photosynthetic activity for that location in the landscape throughout the year. In environments where the ground cover senesces completely the minimum EVI is related to the woody vegetation canopy component.
For areas that support vegetation (i.e. not salt lakes and open water bodies) the trend in the mean provides an estimate of how total photosynthetic activity is changing from year to year. The trend in the annual mean is the most robust of all three trend data analyses because it incorporates all 23 points in the time series for each year i.e. every point in the time series. This also makes it a less sensitive tool than the other two trends, but also more robust as an indicator of change.
Table 1: International Standards Organisation class and equivalent common name conversion table for a subset of the land cover types.
|Common Name||ISO descriptor|
|Closed Tussock Grassland||Tussock Grasses (Closed)|
|Open Tussock Grassland||Tussock Grasses (Open)|
|Sparse Tussock Grassland||Tussock Grasses (Sparse)|
|Very Sparse Tussock Grassland||Tussock Grassland (Scattered)|
|Open Hummock Grassland||Hummock Grasses (Open)|
|Sparse Hummock Grassland||Hummock Grasses (Sparse)|
|Very Sparse Grassland||Graminoids (Scattered)|
|Sparse Grassland||Graminoids (Sparse)|
|Open Sedgeland||Sedges (Open)|
|Dense Shrubland||Shrubs (Closed)|
|Open Shrubland||Shrubs (Open)|
|Sparse Shrubland||Shrubs (Sparse)|
|Very Sparse Shrubland||Shrubs (Scattered)|
|Open Saltbush Shrubland||Chenopods (Open)|
|Sparse Saltbush Shrubland||Chenopods (Sparse)|
|Very Sparse Saltbush Shrubland||Chenopods (Sparse)|
|Closed Forest||Trees (Closed)|
|Open Forest||Trees (Open)|
|Open Woodland||Trees (Scattered)|
Glossary of International Standards Organisation terms
Table 2: Glossary of International Standards Organisation terms.
|DLCM Class||Description of equivalent or commonly used cover and vegetation classes|
|Alpine||Vegetated and rocky areas above the tree-line where the vegetation is covered by snow for several months each year. Includes native tussock grasses and forbs and scattered low shrubs.|
|Aquatic vegetation||Vegetated areas associated with wetlands and ponds and rivers includes trees, shrubs and grass-like growth forms.|
|Bare areas||Non-vegetated areas where vegetated cover is absent or could not be detected. Includes naturally bare areas and areas made bare through land management practices.|
|Chenopods||Chenopods are native shrubs in the family Amaranthaceae, formerly Chenopodiaceae. The dominant genera include include Sclerolaena, Atriplex (salt bush), Maireana (blue bushes, cotton bush), Chenopodium and Rhagodia.|
|Cropping||Production of plant species usually managed as a monoculture for food and/or fibre. Native vegetation has largely been replaced by introduced species through clearing, and sowing of new species, the application of fertilisers or the dominance of volunteer species. Dryland and irrigated cropping are defined. Includes production of annual and perennial species.|
|Forbs||Forbs are native and non-native herbaceous flowering plants that are not graminoids. Forbs represent a group of plant communities with broadly similar growth form (e.g. grasses, sedges and rushes). (see gramminoids and sedges)|
|Grassland||Other type of grassland not described by Hummock and Tussock grasslands (may be native or non-native).|
|Graminoids||Graminoids are native and non-native non-woody plants with narrow leaves growing from the base including the "true grasses", of the Poaceae (or Gramineae) family, and sedges (Cyperaceae) and rushes (Juncaceae). (see forbs)|
|Hummock||A coarse xeromorphic native grass with a mound-like form often dead in the middle. The main genera are Triodia, Plectrachne and Zygochloa. Forms extensive areas either as the dominant growth form or in association with shrubs and trees.|
|Mining||A cover class where the vegetation is usually removed to extract the underlying minerals and rocks.|
|Pastures||Pasture and forage production, both annual and perennial, is based on a significant degree of modification or replacement of the native vegetation. Areas are cultivated or maintained for the production of food for animals, whether harvested or grazed directly. Dryland and irrigated are defined.|
|Sedges||Herbaceous species, usually perennial, with a tufted habit. Includes the plant families Cyperaceae (true sedges) and Restionaceae (node sedges). (see forbs and gramminoids)|
|Shrubs||Woody plants, multi-stemmed at the base (or within about 200 mm from ground level), or, if single-stemmed, less than about 5 m tall; not always readily distinguishable from small trees.|
|Sugar||A cultivated crop comprising tall perennial grass of the genus Saccharum (family Poaceae tribe Andropogoneae). Includes dryland and irrigated practices.|
|Trees||Native and non-native woody plants more than 2 m tall usually with a single stem, or branches well above the base; not always distinguishable from large shrubs.|
|Tussock||Native grass communities and non-native species forming extensive areas dominated by a few species. Family Poaceae with a tufted habit. Tussock communities are usually dominated by particular genera such as Astrebla, Austrodanthonia, Austrostipa, Dicanthium, Eragrostis, Poa, Themeda, Sorghum, Heteropogon, Ophiuros, Oryza, Eragrostis and Spinifex.|
|Water||Land surface water features include rivers, lakes and ponds. Fresh and saline water bodies are defined. Includes permanent and intermittent water features.|
The creation of the Land Cover Map was made possible by funding support from Geoscience Australia, ABARES and Caring for Our Country. These contributions included the provision of data for the initial creation and labelling of the classes, comparative data to revise and validate the land cover classes, detailed validation work and participation in the Dynamic Land Cover Dataset workshop. We thank the following organisations for valuable contributions to the development of the Land Cover Data:
- Australian Collaborative Land Use and Management Program (ACLUMP)
- Australian Collaborative Rangelands Information System (ACRIS)
- Commonwealth Scientific and Industrial Research Organisation (CSIRO)
- Executive Steering Committee for Australian Vegetation Information (ESCAVI)
- National Committee for Land Use and Management Information (NCLUMI)
- Queensland Statewide Landcover And Tree Study (SLATS)
- State and Territory Government Natural Resource Management Departments.
When citing this dataset please cite: Lymburner L., Tan P., Mueller N., Thackway R., Lewis A., Thankappan M., Randall L., Islam A., and Senarath U., (2010), 250 metre Dynamic Land Cover Dataset of Australia (1st Edition), Geoscience Australia, Canberra.
Topic contact: firstname.lastname@example.org Last updated: November 23, 2011