DEA Waterbodies (Landsat)

Geoscience Australia Landsat Waterbodies Collection 3

Version
2.0.0 flat-square
Product ID
ga_ls_wb_3
Program
Digital Earth Australia
Collection
Geoscience Australia Landsat Collection 3
Resource type
Derivative
Published Date
27/01/2022

About

Locate over 300,000 waterbodies across Australia and look back at their changes over three decades with Digital Earth Australia (DEA) Waterbodies. Monitor critical lakes and dams, including hard-to-reach waterbodies in remote areas and on large properties.

Background

Up-to-date information about the extent and location of surface water in Australia provides us with a common understanding of this valuable and increasingly scarce resource.

This dataset is version 2 of DEA Waterbodies, and represents a reprocessing of DEA Waterbodies version 1 on DEA Collection 3 Water Observations, as well as a few incremental improvements. See the "Details" tab for a discussion of the changes between DEA Waterbodies v1 and v2.  

What this product offers

Digital Earth Australia Waterbodies uses Geoscience Australia’s archive of over 30 years of Landsat satellite imagery to identify where over 300,000 waterbodies are in the Australian landscape and tells us the wet surface area within those waterbodies.

The tool uses a water classification for every available Landsat satellite image and maps the locations of waterbodies across Australia. It provides a time series of wet surface area for waterbodies that are present more than 10% of the time and are larger than 2,700m2 (3 Landsat pixels).

The tool indicates changes in the wet surface area of waterbodies. This can be used to identify when waterbodies are increasing or decreasing in wet surface area.

Applications

  • Understand local through to national-scale surface water dynamics over time and geography
  • Provide supporting information to better manage water across Australia
  • Gain insights into the severity and spatial distribution of drought
  • Identify potential water sources for aerial firefighting during bushfires
  • Get deeper insight into DEA Water Observations data

Related products

Publications

Krause, Claire E.; Newey, Vanessa; Alger, Matthew J.; Lymburner, Leo. 2021. "Mapping and Monitoring the Multi-Decadal Dynamics of Australia’s Open Waterbodies Using Landsat" Remote Sens. 13, no. 8: 1437. https://doi.org/10.3390/rs13081437

Mueller, N., Lewis, A., Roberts, D., Ring, S., Melrose, R., Sixsmith, J., Lymburner, L., McIntyre, A., Tan, P., Curnow, S., & Ip, A. (2016). Water observations from space: Mapping surface water from 25 years of Landsat imagery across Australia. Remote Sensing of Environment, 174, 341–352. https://doi.org/10.1016/j.rse.2015.11.003

Data access

Link to data Download the shapefile via eCat
Digital Earth Australia - Public Data
Dataset technical metadata
Link to maps DEA Maps
Code examples Jupyter notebook
Github repository
Web services Web mapping service
eCat record 146197
DOI dx.doi.org/10.26186/146197
Product ID ga_ls_wb_3
CMI RESTful node ID 693
Security classification Unclassified
Update frequency monthly

Access notes

DEA Maps

To explore DEA Waterbodies on the interactive DEA Maps platform, visit the link below:

https://maps.dea.ga.gov.au/#share=s-4RjD9N7swBCZSkXRJJBxOLzVZyK

and dismiss the overlay. DEA Waterbodies will already be loaded on the map.

DEA MAps with DEA Waterbodies loaded

 

To add DEA Waterbodies to DEA Maps manually:

1) Visit DEA Maps.

2) Click 'Explore map data'.

3) Select 'Inland water' > 'DEA Waterbodies'.

4) Select 'DEA Waterbodies (version 2)' and click 'Add to the map'.

See Waterbodies user guide to discover how to get the most out of the DEA Maps presentation of DEA Waterbodoes including how to explore time series, and how to download the data from a particular time series.

DEA Maps loaded with DEA Waterbodies and showing Feature Information

Technical information

The DEA Waterbodies product is comprised of two key components:

- a polygon dataset of automatically mapped waterbody outlines, and

- a csv time series for each polygon capturing the surface area of water within each polygon at every available, clear Landsat observation. 

Producing DEA Waterbodies

DEA Waterbodies is a polygon-based view of DEA Water Observations (DEA WO), derived through the automatic processing of DEA WO to identify the outlines of persistent waterbodies across Australia (Figure 1). 

Flow diagram outlining the steps taken to produce DEA Waterbodies polygons
Figure 1: Flow diagram outlining the steps taken to produce DEA Waterbodies v2 polygons

For a detailed discussion of the methods used to produce DEA Waterbodies v1, refer to Krause et al. 2021

DEA Waterbodies v2

The key difference between DEA Waterbodies v1 and v2 is the underlying satellite imagery used to derive the polygons, and to generate the accompanying csvs. DEA Waterbodies v1 was produced on the DEA Water Observations from Space product, which was derived from DEA Surface Reflectance NBART (Landsat) collection 2 data. This dataset had a resolution of 25m. 

DEA Waterbodies v2 has been reprocessed on DEA Water Observations, which has been run on Landsat 5 TM Level-2 Data Products - Surface Reflectance / Landsat 7 ETM+ Level-2 Data Products - Surface Reflectance / Landsat 8 OLI/TIRS Level-2 Data Products - Surface Reflectance collection 3 data, which has a pixel resolution of 30m. The reprocessing of DEA datasets to 30m resolution required that DEA Waterbodies polygon generation be re-run to re-map each waterbody using the new pixel resolution.

DEA Waterbodies v2 differs from v1 in a few additional key areas: 

  • v2 polygons have a minimum size of 2,700 m2, while v1 polygons have a minimum size of 3,125 m2 ;

  • There are new waterbodies in v2 not present in v1; 

  • There are old waterbodies in v1 that are not present in v2, mainly including very small or rarely full polygons; and 

  • There are waterbodies in both datasets for which the outlines have changed between v1 and v2. 

Change in minimum polygon size

A change in the underlying pixel size necessitated a re-evaluation of the minimum polygon size. In v1, the minimum polygon size was 3,125m2, equating to 5 Landsat collection 2 pixels. 

In v2, the size has been lowered slightly to 2,700m2, which equates to 3 Landsat collection 3 pixels. 

This change has resulted in the inclusion of some smaller waterbodies that were not mapped in v1 (Figure 2). 

Comparison of the number of smaller waterbodies in v1 compared to v2. The second panel shows some waterbodies that have been included in v2 that were missed in v1.
Figure 2: Comparison of the size distributions of DEA Waterbodies v1 and v2. a) Size distribution for polygons smaller than 1km2. b) Small waterbodies identified in v2 that were not included in v1.

Waterbody polygons manually curated in v2

Our automated waterbody polygon detection produces subpar results for large, very rarely filled waterbodies. This is particularly true of the large salt lakes in South Australia, where our method produces thousands of smaller polygons instead of the single encompassing polygon that is typically used to map these salt lakes. These subpar results come from a combination of elevation and satellite imaging effects. To mitigate these effects, we replaced the most complex large waterbodies with their counterparts in the Surface Hydrology Polygons (Regional) dataset: 

  • Kati Thanda-Lake Eyre (North) 
  • Kati Thanda-Lake Eyre (South) 
  • Lake Torrens 
  • Lake Frome / Munda 
  • Lake Gairdner 
  • Lake Blanche 
  • Lake Everard 

As an example, Kati Thanda is a particularly complex polygon when mapped using automated methods, resulting in 3,118 polygons ranging from 900 m2 to 4,609 km2. Kati Thanda and Lake Everard in v1 and v2 are shown in Figure 3. 

Large, complex salt lakes in v1 were swapped with polygons from the Surface Hydrology dataset in v2
Figure 3: Complex polygons replaced for large salt lakes in South Australia. Original polygons are shown in red in the foreground. The polygons that replaced these are shown in the background in blue. a) Focus on the northern edge of Kati Thanda, SA, showing the complexity of the automatically detected polygons. b) Lake Everard, SA.

Polygon names

DEA Waterbodies polygons are named using a geohash as the unique identifier for each polygon. A geohash is a representation of the lat/lon coordinates of the centre of each polygon, mapped into a shorter character string. Each polygon's geohash can be converted back to a lat/lon pair to make it easy to locate a waterbody from its geohash alone. 

Polygon names/geohashes are not maintained between v1 and v2 of DEA Waterbodies. As in v1, v2 waterbody polygons are named according to the centroid of each polygon, which may have moved between versions. Note that characters at the end of a geohash string represent increasing precision in the accompanying lat/lon coordinate pair, so while the geohash is not directly maintained between versions, it is likely that they will be similar, with only the last few characters varying due to slight differences in the polygon centroids.

In order to prevent polygon identifiers being mixed up between versions, we have introduced explicit version numbers to each unique ID. This has also been applied retrospectively to v1 so that all DEA Watebody polygons are now named with a geohash and a version number. For example, in v1, Kati Thanda was originally r4ctk0hzm. In v1.1, the long-term release of v1, Kati Thanda was r4ctk0hzm_v1. In v2 Kati Thanda is r4ctum36x_v2

Accuracy and limitations

For a full discussion of the accuracies and limitations of DEA Waterbodies, please refer to Krause et al. 2021

Inaccuracies inherited from DEA Water Observations (WO)

Many of the inaccuracies and limitations of the waterbody analysis are inherited from WO, with this product a reanalysis and mapping product built off the WO datasets. WO has a number of known limitations, and these manifest as misclassified waterbodies within this analysis. WO uses the spectral signature of water to classify wet pixels, and is known to be suboptimal in locations where water and vegetation are mixed. This includes locations such as rivers with vegetated riparian zones and vegetated wetlands. The effect of this can be seen by the discontinuity of narrower river features identified within this analysis, and an under representation of water within vegetated wetlands, such as the Macquarie Marshes, NSW.

Other known WO limitations have been limited through the filtering processes used to produce the map of waterbodies. Issues with mixed water and vegetation pixels around features like small farm dams have been avoided by limiting the size of mapped waterbodies to at least five Landsat pixels. Misclassification of water in deep shadows in high density cities has been handled by removing any waterbody polygons identified within CBDs. Intermittently misclassified features, which return valid results only a handful of times over the 32 year study period, are also filtered out by testing for the number of valid observations returned for each pixel.

Despite this, some errors remain in the final waterbodies dataset. Steep terrain shadows present a known difficulty for the WO classifier, due to the shadows produced. While WO has attempted to mitigate this issue, some misclassification remains. We have not specifically attempted to address these errors within this workflow, and as such, a negligible number of the identified waterbodies may in fact be artifacts caused by terrain shadow. The signal to noise ratio over deeper water has also not been addressed here, and may result in some pixels missing from the centre of deeper waterbodies, resulting in doughnut-shaped mapped polygons. Similarly, different water colours may interfere with the decision-tree classifier, resulting in very turbid or coloured waterbodies being misclassified.

The automatic cloud masking algorithm used in this analysis can misclassify bright, white sands seen on the bottom of some waterbodies as clouds. This issue is particularly problematic where these bright sands are only exposed when the waterbody begins to empty, resulting in the bright sands being seen inconsistently over time. It is very difficult to accurately cloud mask these sands, as they are seen in some scenes but not others, in the same way that clouds come and go between scenes. In this version of DEA Waterbodies, we have not addressed this issue, and note it as a limitation that results in short or missing timeseries; the sands are incorrectly classified as cloud, and the scene is thrown out as being unsuitable, resulting in very few ‘clear’ scenes. We hope to address this issue in the next release of the DEA Waterbodies product.

For a full discussion of the limitations and accuracy of WO, see Mueller et al. (2016).

Relevant websites

References

Krause, Claire E.; Newey, Vanessa; Alger, Matthew J.; Lymburner, Leo. 2021. "Mapping and Monitoring the Multi-Decadal Dynamics of Australia’s Open Waterbodies Using Landsat" Remote Sens. 13, no. 8: 1437. https://doi.org/10.3390/rs13081437

Mueller, N., Lewis, A., Roberts, D., Ring, S., Melrose, R., Sixsmith, J., Lymburner, L., McIntyre, A., Tan, P., Curnow, S., & Ip, A. (2016). Water observations from space: Mapping surface water from 25 years of Landsat imagery across Australia. Remote Sensing of Environment, 174, 341–352. https://doi.org/10.1016/j.rse.2015.11.003

Lineage

This product builds upon DEA Water Observations, details of which are available here.

The code used in the development of this product is available on GitHub.

Data sources

Major algorithms

Schema / spatial extent

DEA Waterbodies (Landsat)

Update frequency monthly
Temporal extent
Coordinate reference system Australian Albers / GDA94 (EPSG: 3577)
Cell size X 30.00
Cell size Y 30.00

Example images

Conceptualisation of the time series extraction for each polygon. At each individual time step, the percentage surface area of water within each waterbody polygon (outlined in red) was calculated, producing a time series (d) of the change in surface water
Time series of the change in the percentage of total surface area observed as water for waterbody ID r6f07ug9r_v1. The corresponding false colour imagery is shown for three time steps, showing the relationship between the time series and the raw imagery.
National-scale seasonal wet area statistics. See Krause et al. (2021) for full details.

Owner

Commonwealth of Australia (Geoscience Australia)

Principal contributors

Vanessa Newey

Subject matter experts

Vanessa Newey

License

CC BY Attribution 4.0 International License

Rights statement

© Commonwealth of Australia (Geoscience Australia) 2022. Creative Commons Attribution 4.0 International License.

Acknowledgments

This work was carried out in collaboration with NSW Department of Planning, Industry and Environment.