DEA Water Observations (Sentinel-2 NRT)

DEA Water Observations (Sentinel-2 NRT)

Geoscience Australia Sentinel-2 Near Real Time Water Observations Collection 3

Version:

1.1.6 (Latest)

Product types:

Derivative, Raster

Time span:

14 Mar 2023 – Present

Update frequency:

Daily

Product ID:

ga_s2_wo_provisional_3

../../../_images/S2_WO_Provisional_Gulf_20230320_0.png

Provisional product

This is a provisional product, meaning it has not yet passed quality control and/or been finalised for release.

About

This is a rapid, provisional, product. It has not been validated and is of unknown accuracy, however it may provide useful information about the general extents of surface water soon after an image is acquired.

Access the data

For help accessing the data, see the Access tab.

Access the data on AWS

Get the data online

Code examples

Code sample

Web Services

Get via web service

Key details

Parent product(s)

Sentinel 2A and 2B NBART and Observational Attributes

Collection

Geoscience Australia Sentinel-2 Collection 3

DOI

10.26186/146257

Licence

Creative Commons Attribution 4.0 International Licence

Cite this product

Data citation

Geoscience Australia, 2022. DEA Water Observations (Sentinel-2 NRT; Provisional). Geoscience Australia, Canberra. https://dx.doi.org/10.26186/146257

Publications

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

Background

Understanding the amount of surface water in the landscape is important for many applications including water management, wetlands, water connectivity, surface-groundwater relationships and flood mapping. This can lead to more effective emergency management and risk assessment.

This product is the implementation of the DEA Water Observations (DEA WO) product (previously known as Water Observations from Space, or WOfS) on the Geoscience Australia Sentinel-2 Near Real Time surface reflectance product.

The Provisional Digital Earth Australia Water Observations (Sentinel-2) product shows where surface water was observed by the Sentinel-2A and Sentinel-2B satellites on any particular day over the most recent 3 months. The surface water observations are derived from Geoscience Australia Sentinel-2 Near Real Time surface reflectance imagery for all of Australia. The provisional, Near Real Time product is available for a rolling window of the most recent three months of data, and is produced within 24 hours of the satellite passing over an area.

What this product offers

The Provisional Digital Earth Australia Water Observations product shows the extent of surface water in a corresponding Sentinel-2 scene, along with the degree to which the scene was obscured by data quality issues including; cloud, cloud shadows, and where sensor problems cause parts of a scene to not be observable. The Water observations are based on the WOfS algorithm which underpins all of the Digital Earth Australia surface reflectance water products.

This is a rapid, provisional, product. It has not been validated and is of unknown accuracy, however it may provide useful information about the general extents of surface water soon after an image is acquired.

Applications

The Digital Earth Australia Near Real Time Water Observation product is provided as a provisional (not validated, with unknown accuracy) source of information on water in the landscape to help inform emergency management and decision makers on recent and evolving emergency situations. The product can be used to estimate the area of surface water present in the corresponding satellite scene, and can be used for several water monitoring applications. Uses of the individual Water Observations (WOs) includes:

  • flood extent

  • amount of water in water bodies, major rivers and the coastal zone.

As the WOs are separated from the derived statistics of the associated DEA Water statistical products, the WOs are most useful for performing analyses requiring the investigation of surface water extent for particular times rather than over long term time series.

Technical information

Digital Earth Australia (DEA) Water Observations (Sentinel-2) Provisional is a dataset indicating areas where surface water has been observed using the Geoscience Australia Sentinel-2 surface reflectance imagery catalogue. The Near Real Time version of this product includes observations from the most recent three months of data collected by the Sentinel-2 satellites. Near Real Time products are created with predicted corrections for atmospheric conditions and ground reflectivity. It covers all of mainland Australia and Tasmania but excludes off-shore Territories. The dataset is updated automatically as each new Sentinel-2 scene is acquired and processed to Near Real Time Analysis Ready Data (ARD) state.

The Water Observations from Space (WOfS) Water detection algorithm is used to generate a pixel wise classification of Sentinel-2A and Sentinel-2B scenes. The output of this algorithm is a classification with the bit meanings described in Table 1. Because of the way the current version (Version 3.1.6) of the WOfS algorithm operates multiple bits can be set simultaneously for each pixel. Hence the value of a pixel in an observation can be X AND Y AND Z, etc, hence values can range from 0 to 255.

Table 1: The table below describes the meaning of each bit set per pixel. Where multiple factors impeding a clear observation are detected all the respective bits will be set. For example a value of 136 indicates water (128) AND terrain shadow (8) were observed for the pixel.

Bit

Flagging

Decimal Value

Description

0

no data

1

Pixel masked out due to NO_DATA in NBART source, 0 = valid data in NBART

1

contiguity

2

Pixel masked out due to lack of data contiguity

2

solar incidence

4

Pixel masked out due to solar incidence > 10 degrees

3

terrain shadow

8

Pixel masked out due to terrain shadow

4

high slope

16

Pixel masked out due to high slope

5

cloud shadow

32

Pixel masked out due to cloud shadow

6

cloud

64

Pixel masked out due to cloud

7

water

128

Water detected

Full details of the original algorithms and features of DEA Water Observations can be found in the Water Observations from Space paper by Mueller et al. (2016). https://doi.org/10.1016/j.rse.2015.11.003

Lineage

The DEA Water Observations dataset is generated using the following steps:

Imagery of the earth’s surface is collected by the Sentinel-2A and Sentinel-2B satellites as a raster based product. Digital Earth Australia receive this data and conduct atmospheric and Bidirectional Reflectance Distribution Function corrections to produce Geoscience Australia Sentinel-2 Analysis Ready Data to predicted (NRT) standard.

Once the input data has been processed to ARD the Water Observations from Space (WOfS) algorithm is applied, including cloud and shadow masking, on a per-granule basis and indexed into the Data Cube for further processing.

Processing steps

1. Surface Reflectance Calculation (NBAR + Terrain Illumination Correction)

Calculate a standardised optical surface reflectance using robust physical models to correct for variations in image radiance values due to atmospheric properties, sun and sensor geometry and accounts for the directional reflectance properties of the surface. The process also accounts for the underlying surface topography via a process known as terrain illumination correction.

The process of producing standardised optical surface reflectance is known as Nadir corrected Bi-directional reflectance distribution function Adjusted Reflectance (NBAR) coupled with a Terrain illumination correction.

This enables comparison of imagery acquired at different times, in different seasons, in different geographic locations over varying surface topography.

Algorithm details can be found here: https://doi.org/10.1016/j.rse.2012.06.018

2. Water Observations from Space Detection Algorithm

The water detection algorithm used to detect water from each observed pixel is based on a statistical regression tree analysis of a set of normalised difference indices and corrected band values. The regression is based on a set of water and non-water samples created by visual interpretation of 20 Landsat scenes from across Australia. The sample locations, indicated below, ensure that the logistic regression is based on the full geographic range of conditions experienced in Australia.

The regression analysis determined a set of best indices and bands for the analysis and the associated thresholds in each component to derive a final classification tree, producing a water/non-water classification for pixel in the Data Cube. The final water classification for each pixel is modified by Pixel Quality (see associated PQ product information) and terrain. Once the water algorithm has completed its process, the water detection for a pixel through time is combined to produce a total number of water observations for each pixel. This is compared to a total number of clear observations for the same pixel, derived from the PQ analysis. The ratio is expressed as a percentage water recurrence.

Confidence that a pixel depicted as having had water detected at some time is calculated by a Confidence Layer. The layer is computed by combining a set of confidence factors using a weighted sum approach, with the weightings derived by logistic regression.

References

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

Accuracy

This is a rapid, provisional, product. It has not been validated and is of unknown accuracy. The accuracy of the original Water Observations from Space (WOfS) algorithms was assessed using an independent set of 3.4 million validation points using Landsat imagery only. A similar validation has not yet been performed on this Sentinel-2 product, and the accuracy described below may serve as only a rough indicator.

Landsat WOfS validation points were identified based on visual interpretation of Landsat imagery within 20 test areas across Australia The points were identified in the same locations as the training data, but were selected from different years (i.e. imagery from one set of years was used to generate points to train the algorithm, and imagery from a separate set of years was used to generate the points that were used evaluate the accuracy of the algorithm).

The classification has an overall accuracy of 97%. Areas identified as water within the accuracy assessment data are being correctly identified 93% of the time and are being misclassified as not water 7% of the time. These errors of omission typically occur along rivers, small waterbodies and swamps where the presence of both water and vegetation within the pixel leads a failure to identify water. This means that the DEA Water Observations product is likely to underestimate the extent of water in locations that contain mixed water and vegetation pixels. As a consequence of this the product may not be fit for applications that require information about the inundation characteristics of vegetated wetlands, small farm dams, and rivers less than 50 metres wide.

Water can be incorrectly detected by the classification algorithm in areas where steep terrain or tall buildings cause frequently shaded pixels. These errors of commission are occurring in 8% of samples used to evaluate the accuracy of the classification. This means that the product may overestimate the amount of water in locations that are adjacent to steep terrain or in dense urban areas. Terrain masks and urban masks were used in the confidence layer to reduce this overestimation, however some residual errors remain. As a consequence of this the product may not be fit for applications that require information about the inundation characteristics of urban areas or locations adjacent to steep terrain.

In addition to the limitations of the classification algorithm, the satellite observation frequency also introduces limitations to the product. The product is likely to be underestimating the extent of inundation for infrequent flood events because the 5 day revisit frequency (best case scenario notwithstanding the possibility of cloud obscuring the floodwaters) will potentially fail to observe the flood peak. This is an intrinsic limitation of the observation strategy. As a consequence of this limitation, the product is not suited to applications that require a. the identification of a ‘maximum extent of inundation’ line, or b. detailed information about the extent of infrequent flood events.

Limitations

Observation of Earth by the satellites used for this service depends on clear skies. Furthermore, the satellites do not observe all places every day. Each of the Sentinel-2 satellites, which are the basis for this service, view a given 290 kilometre wide strip of Australia only once every 10 days. The observations show only what was visible on the day of the satellite pass. As a result, not all historical floods will have been observed by satellite.

The automated surface water detection algorithm, which has been developed by Geoscience Australia, can sometimes mistakenly label large buildings; cloud shadow; large uniform black tarpaulins; or snow as “water”. The algorithm is designed to locate large areas of water and as a result may miss small water bodies.

The algorithm also uses ‘fmask’ for cloud masking, this method is known to have limitations in correctly identifying clouds, particularly across bright substrates like urban areas, beaches and salt-pans.

The satellite archive used for this service is of limited duration (most recent 3 months), and subject to the cloud and repeat coverage restrictions noted above. In addition, Australia is subject to wide variations in weather and climate which can lead to lengthy periods where areas are not observable.

This product is produced at 10 metre resolution. This is achieved by resampling the 20 metre Sentinel-2 spectral bands to 10 metre resolution using “bilinear” resampling. This is likely to affect the ability of the product to correctly classify narrow areas of water (e.g. less than 20 metres in width).

Access the data

Explore data availability

Learn how to use the DEA Explorer

Get the data online

Learn how to access the data via AWS

Code sample

Learn how to use the DEA Sandbox

Get via web service

Learn how to use DEA’s web services

Use constraints

The DEA WO (Sentinel-2) Provisional product has not been validated, and has unknown accuracy. It shows the general extent of surface water in the corresponding Sentinel-2 satellite image. These extents should not be used for safety of life decisions without confirming the information with other sources.

How to view the data in a web map

To view and access the data interactively:

  1. Visit DEA Maps.

  2. Click Explore map data.

  3. Select Inland water > DEA Water Observations > DEA Water Observations (Sentinel-2, Provisional).

  4. Click Add to the map, or the + symbol to add the data to the map.

Old versions

No old versions available.