Water Observations from Space

Date modified: 24 March 2021

The WOfS product suite is currently a beta product generated from provisional Landsat 8 satellite observations, and will be operationalised in the second quarter of 2021 to use full Landsat Collection 2 data.

Product overview

Background

Water Observations from Space (WOfS) is a service that draws on satellite imagery to provide historical surface water observations of the whole African continent. WOfS allows users to understand the location and movement of water present in the African landscape by giving them an improved understanding of where water is usually present; where it is seldom observed; and where inundation of the surface has been observed by satellite.

They are generated using the WOfS classification algorithm on Landsat 8 satellite data. There are several WOfS products available for the African continent, as listed below:

Product Type

Description

WOfS Feature Layer (WOFL)

WOfS slices generated per scene

WOfS Annual Summary

The ratio (%) of wet to clear observations from each calendar year

WOfS All-Time Summary

The ratio (%) of wet to clear observations over all time

  • WOfS Feature Layer (WOFL): Individual water-classified images are called Water Observation Feature Layers (WOFLs), and are created from the input satellite data. There is one WOFL for each satellite dataset processed for the occurrence of water.

  • WOfS Annual/All-Time Summary: The percentage of time a pixel was classified as wet. This requires:

    • Total number of clear observations for each pixel: the number of observations that were clear (no cloud or shadow) for the selected time period. The classification algorithm then assigns these as either wet, or dry.

    • Total number of wet observation for each pixel: the number of observations that were clear and wet for the selected time period.

The WOfS summaries are calculated as the ratio of clear wet observations to total clear observations, expressed as a percentage.

\[\text{WOfS Summary} = \frac{\text{Number of Wet Observations}}{\text{Number of (Wet + Dry) Observations}} \times 100\]

Specifications

There are several datasets that are available as part of the WOfS suite. Relevant coverage and metadata can be viewed on DE Africa Metadata Explorer:

WOfS Feature Layer

Table 1: WOfS Feature Layer product specifications

Specification

Product name

WOfS Feature Layer

Cell size - X (metres)

30

Cell size - Y (metres)

30

Coordinate reference system

EPSG: 6933

Temporal resolution

16 days

Temporal range

2013-03-19 – 2019-06-30

Parent dataset

Provisional Landsat 8 Collection 2

Update frequency

Not applicable for beta products

Table 2: WOfS Feature Layer measurements

Band ID

Description

Value range

Data type

No data value

water

WOFL water

0 - 255

uint8

1

The WOFL measurement water uses bit flags to allocate terrain characteristics to each pixels. Bit flags assign a unique decimal value to each characteristic. A pixel can hold multiple characteristics by summing the decimal values of each associated bit flag.

Table 3: WOfS Feature Layer bit flags

Bit

Flagging

Decimal Value

Description

0

no data

1

1 = pixel masked out due to NO_DATA in source, 0 = valid data

1

non-contiguity

2

Lack of data contiguity

2

sea

4

Sea

3

terrain shadow / low solar angle

8

Terrain shadow or low solar angle

4

high slope

16

High slope

5

cloud shadow

32

Cloud shadow

6

cloud

64

Cloud

7

water observed

128

Water present

For example, a water value of 136 indicates water (128) AND terrain shadow / low solar angle (8) were observed for the pixel, whereas a value of 144 would indicate water (128) AND high slope (16).

WOFLs are useful for identifying the presence of water at a certain point in time, or over short durations (less than one year). For annual or historical data, users can access the pre-calculated summary products detailed below.

WOfS Annual Summary

Table 4: WOfS Annual Summary product specifications

Specification

Product name

WOfS Annual Summary

Cell size - X (metres)

30

Cell size - Y (metres)

30

Coordinate reference system

EPSG: 6933

Temporal resolution

Annual

Temporal range

2017 – 2019

Parent dataset

WOfS Feature Layer

Update frequency

Not applicable for beta products

Table 5: WOfS Annual Summary measurements

Band ID

Description

Value range

Data type

No data value

count_wet

How many times a pixel was wet

1 - 65535

uint16

-1

count_clear

How many times a pixel was clear

1 - 65535

uint16

-1

frequency

Frequency of water detection at a location

0 - 1

float32

NaN

WOfS All-Time Summary

Table 6: WOfS All-Time Summary specifications

Specification

Product name

WOfS All-Time Summary

Cell size - X (metres)

30

Cell size - Y (metres)

30

Coordinate reference system

EPSG: 6933

Temporal resolution

One summary for entire temporal range

Temporal range

2013-03-19 – 2019-06-30

Parent dataset

WOfS Feature Layer

Update frequency

Not applicable for beta products

Table 7: WOfS All-Time Summary measurements

Band ID

Description

Value range

Data type

No data value

count_wet

How many times a pixel was wet

1 - 65535

uint16

-1

count_clear

How many times a pixel was clear

1 - 65535

uint16

-1

frequency

Frequency of water detection at a location

0 - 1

float32

NaN

All products in the WOfS suite have the same geographic extent. This is shown in Figure 1; data is available for the regions shaded in blue. Specific temporal and geographic extents can be explored as an interactive map on the Digital Earth Africa Metadata Explorer. Different WOfS products can be selected from the horizontal dropdown menu at the top of the page.

Figure 1: Landsat WOfS suite geographic extent

S-2 GeoMAD data extent

Processing

The Water Observations from Space Detection Algorithm as described in Mueller, 2016, is applied to the Landsat data to create the available WOfS products.

Media and example images

Image 1: Tagrin Bay, Sierra Leone. WOfS All-Time Summary.

Colours indicate the percentage of times water was detected. Red is “rarely water”, green is “often water”, and blue is “always water”.

Credit: U.S. Geological Survey Landsat data was used in compiling this image.

WOfS All-Time Summary over Tagrin Bay.

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

License

CC BY Attribution 4.0 International License

Acknowledgments

The WOfS algorithms incorporated in this product are the work of Norman Mueller, Geoscience Australia, and Dr Dale Roberts, Australian National University.

Data Access

Amazon Web Services S3

The DE Africa WOfS product suite is available in AWS S3 thanks to the Public Dataset Program.

Table 8: AWS data access details

AWS S3 details

Bucket ARN

arn:aws:s3:::deafrica-data

Product names

ga_ls8c_wofs_2, ga_ls8c_wofs_2_annual_summary, ga_ls8c_wofs_2_summary

The bucket is located in the region us-west-2 (Oregon).

The following file path convention applies to WOFLs:

usgs/pc2/ga_ls8c_wofs_2/<path>/<row>/<year>/<month>/<day>/

Annual Summaries omit path, row, month and day.

usgs/pc2/ga_ls8c_wofs_2_annual_summary/<year>/

The All-Time Summary additionally drops year.

usgs/pc2/ga_ls8c_wofs_2_summary/

Table 9: AWS file path convention

File path element

Description

Example

product name

ga_ls8c_wofs_2, ga_ls8c_wofs_2_annual_summary, or ga_ls8c_wofs_2_summary

ga_ls8c_wofs_2

path number

Ranges from 157 to 206.

165

row number

Ranges from 052 to 072.

052

year

Year the data was collected

2019

month

Month of the year the data was collected (with leading zeros)

03

day

Day of the month the data was collected (with leading zeros)

28

OGC Web Services (OWS)

This product is available through DE Africa’s OWS.

Table 10: OWS data access details.

OWS details

Name

DE Africa Services

Web Map Services (WMS) URL

https://ows.digitalearth.africa/wms?version=1.3.0

Web Coverage Service (WCS) URL

https://ows.digitalearth.africa/wcs?version=2.1.0

Layer names

ga_ls8c_wofs_2, wofs_2_annual_summary_frequency, wofs_2_summary_frequency

Digital Earth Africa OWS details can be found at https://ows.digitalearth.africa/.

For instructions on how to connect to OWS, see this tutorial.

Open Data Cube (ODC)

The WOfS suite of datasets can be accessed through the Digital Earth Africa ODC API, which is available through the Digital Earth Africa Sandbox.

ODC product name: ga_ls8c_wofs_2, ga_ls8c_wofs_2_annual_summary, ga_ls8c_wofs_2_summary

Table 11: WOFL ODC band names

Band name

Alternative names

water

The Annual Summary and the All-Time Summary have the same band names in the ODC.

Table 12: WOfS Annual/All-Time Summary ODC band names

Band name

Alternative names

count_wet

wet

count_clear

clear

frequency

freq

For examples on how to use the ODC API, see the DE Africa example notebook repository.

Technical information

Algorithm

The Water Observations from Space Detection Algorithm uses a decision tree method using both spectral band measurements and derived indices as input datasets. It also utilised several ancillary datasets, including slope.

An illustration of the decision tree is shown in Figure 2.

Figure 2: WOfS Detection Algorithm decision tree. Tree branches are shown in green with endpoint for water and not-water displayed as blue and red respectively. Each branch indicates the variable used to split and the resulting balance of water and not-water samples created by the split. Source: Mueller, 2016

The decision tree underlying the WOfS algorithm.

Validation method

The Digital Earth Africa validation task team has many years’ of combined experience validating satellite-derived maps in different regions of Africa, and a long history of collaborating with a wide variety of stakeholders.

To validate WOfS data, the continent was divided into seven Agro-Ecological Zones (AEZ). Analysts from within each regional geospatial organization labelled a large set of sample points using image interpretation of satellite and aerial imagery. Analysts applied their regional and local knowledge to help define class labels and interpret any difficult features, creating a validation dataset that is both accurate and fit for purpose. To ensure WOfS is as accurate as possible, it was validated against a range of data points. 2900 sample points were generated, covering the African continent, including the main islands.

Typically, water classifiers are adept at mapping large, open water bodies — so for this exercise, water features with an area of more than 100 square kilometres were masked out. This ensured that analysis remained focused on areas that are more challenging to map, such as small water bodies with different colours, depths and surrounding environments.

Stratified random sampling was then performed to select locations with different water occurrences and waterbody types. By focusing the sample on the more difficult-to-map areas, this sample scheme allowed the Digital Earth Africa team to understand the limitations of WOfS, and meant that WOfS could be compared to other available datasets. It also resulted in overall accuracy for this exercise appearing lower than expected, since the ‘easy to map’ areas had been removed from the sample design.

Next, analysts labelled each sample point using a visual interpretation of sample points provided by online tool, Collect Earth Online (CEO). CEO allowed multiple analysts to assess points as water or non-water using satellite and aerial images. The labelled sample points were then compared with the WOfS map, determining where the WOfS map service agrees — or disagrees — with the validation dataset.

The sampling design is independent of the WOfS classification, so this validation dataset can be used to provide quantitive comparisons with products such as future versions of WOfS, or other existing surface water maps.

Validation results

At a continental-scale, WOfS is able to accurately identify about 80% of the labelled water features (Figure 3). 94% of its water classifications are correct (Figure 4, left).

At the AEZ level, WOfS performs outstandingly in Eastern, Sahel and Northern AEZs with more than 85% of reference water features in these zones being correctly identified as water (Figure 4, right). The reliability of this classification in the Eastern zone is also very high at more than 96%; and the WOfS classification in all seven AEZ proved to be reliable, with more than 84% user accuracy (Figure 4, left). The Western AEZ is a challenging zone with high cloud coverage and a wet climate, 71.3% of reference water features have been correctly identified as water in WOfS product, and 97.4% of these water features are actually water.

Figure 3: Overall WOfS accuracy, split by Agro-Ecological Zone.

WOfS overall accuracy mapped by Agro-Ecological Zone.

Figure 4: WOfS user’s accuracy (left) and producer’s accuracy (right), split by Agro-Ecological Zone.

WOfS user's and producer's accuracy mapped by Agro-Ecological Zone.

Limitations and caveats

During the WOfS validation assessment, a number of issues associated with input data, validation method and the WOfS algorithm were detected. The validation results should be interpreted with the following caveats:

  • Spatial resolution: The WOfS product is based on 30 m resolution Landsat imagery while the validation data is produced using 10 m resolution Sentinel-2 imagery. WOfS has trouble in areas with mixed pixels (where a pixel covers both water and upland). These areas tend to be on edges of lakes and in wetlands where there is a mix of water and vegetation. The Sentinel-2 imagery can identify these edges at a higher resolution than the current Landsat WOfS product.

  • WOfS algorithm: A few examples highlight how WOfS misses certain water bodies. This information is critical to improving the WOfS algorithm in future iterations. In some cases, the impact can be mitigated by using a temporal summary of WOfS. A waterbody may be missed in one date of WOfS, but over the course of the year it is mapped as water in other dates and therefore mapped as a waterbody in the annual summary.

  • Environmental factors: There are some errors that are hard to correct for. Sediment, floating vegetation etc. are potentially contributing factors in this type of error by changing the colour of water and making detection hard.

  • Temporal resolution: Reference locations may have been observed on different dates by the Sentinel-2 and the Landsat satellites. If the water extent has changed between the dates, a mismatch in the classification is expected.