Joint Research Centre
Global Surface Water - Data Users Guide (v4)
Background
This document provides a detailed technical description of the datasets that are being made available as
part of the Joint Research Centre's Global Surface Water Dataset. For each of the datasets it provides a
purpose, description, bands and symbology so that users can understand each of the datasets and use
them efficiently and appropriately. Much of the information in this document is taken directly from the
following paper which provides a detailed description of the method that was used to create these
datasets:
Jean-Francois Pekel, Andrew Cottam, Noel Gorelick, Alan S. Belward, High-resolution mapping of global
surface water and its long-term changes. Nature 540, 418-422 (2016). (doi:10.1038/nature20584)
The technical descriptions given here relate to both the data that is available in Google Earth Engine and
to the data that is available for download from the data access section of the website. The data access
section also has ancillary files available which will help when working with the data, such as
symbologies, metadata files and other files. The data that are available in Google Earth Engine are
available in a number of different multi-band assets: one for the mapped products, one for the water
history and one for the metadata datasets.
The datasets that are available are intended to show different facets of the spatial and temporal
distribution of surface water over the last 38 years. Some of those datasets are intended to be mapped
(e.g. the seasonality layer) and some are intended to show the temporal change at specific locations
(i.e. the water history). In addition, there are datasets that provide metadata on the number of
observations and valid observations that were used and these can be used to estimate the confidence
levels in the data. The following sections describe the map products, water history and metadata
datasets.
Map products
Water Occurrence
Purpose:
The Water Occurrence shows where surface water occurred between 1984 and 2021 and provides
information concerning overall water dynamics. This product captures both the intra and inter-annual
variability and changes.
Description (from the paper)
The frequency with which water was present on the surface from March 1984 to December 2021
was captured in a single product called surface water occurrence (SWO). To compute SWO, the
water
detections (WD) and valid observations (VO) from the same months are summed, that is, water
detections and valid observations from March 1984 are added to water detections and valid
observations from March 1985 and so on, such that SWOmonth = ∑WD month / ∑VO month Averaging
the results of all monthly SWOmonth calculations gives the long-term overall surface water occurrence.
The month-by-month time step normalizes occurrence against seasonal variation in the number of valid
observations across the year. Typically, more cloud-free observations (and thus valid observations) are
available during dry seasons than wet. Without monthly weighting, the overall water occurrence (that
is, computed over the full period) would be biased by temporal distribution of the valid observations
(that is, giving more weight to the dry season than to the wet season).
An example of the Water Occurrence dataset is shown below for Mekong river. The permanent water
surfaces (100% occurrence over 38 years) are represented in blue, and areas where water sometimes
occurs are shown in pink through to purple. The paler shades are areas where the water occurs less
frequently.
Bands
The occurrence dataset contains the following band:
Name Data type Description
occurrence unsigned int8 The occurrence value expressed as a percentage
In addition, statistics on the overall number of observations and valid observations that were used to
compute the occurrence (and the other datasets) are provided in the metadata datasets (see the
Metadata datasets section).
Symbology
The occurrence dataset has the following values and symbology. The values from 1 to 100 are discrete.
Value Symbol Colour Label
0 #FFFFFF Not water
1 #FF0000 (1% opacity) 1% occurrence
100 #0000FF (100% opacity) 100% occurrence
255 #CCCCCC No data
Occurrence Change Intensity
Purpose
The Occurrence Change Intensity map provides information on where surface water occurrence
increased, decreased or remained the same between 1984-1999 and 2000-2021 Both the direction of
change and its intensity are documented.
Description (from the paper)
Change in water occurrence intensity between two epochs (16 March 1984 to 31 December 1999, and 1
January 2000 to 31 December 2021) was also produced. This is derived from homologous pairs of
months (that is, the same months contain valid observations in both epochs). The occurrence difference
between epochs was computed for each pair and differences between all homologous pairs of months
were then averaged to create the surface water occurrence change intensity map. Areas where there
are no pairs of homologue months could not be mapped. The averaging of the monthly processing
mitigates variations in data distribution over time (that is, both seasonal variation in the distribution of
valid observations, temporal depth and frequency of observations through the archive) and provides a
consistent estimation of the water occurrence change.
An example of the Occurrence Change Intensity dataset is shown below for the Brahmaputra River.
Increases in water occurrence are shown in green and decreases are shown in red. Black areas are those
areas where there is no significant change in the water occurrence during the 1984 -2021 period. The
intensity of the color represents the degree of change (as a percentage). For example, bright red areas
show greater loss of water than light red areas. Some areas appear grey in the maps, these are locations
where there is insufficient data to compute meaningful change statistics.
Bands
The Occurrence Change Intensity dataset contains the following bands:
Name Data type Description
The absolute difference in the mean
occurrence value between the two epochs
change_ab s unsigned int8 for homologous months
The normalised difference in the mean
occurrence value between the two
change_nor m unsigned int8 epochs for homologous months epoch1-
epoch2 /epoch1+epoch2
Symbology
The Occurrence Change Intensity dataset has the following values and symbology for the change_norm
band. The values from 0 to 200 are discrete.
Value (TIFF) Value (GEE) Symbol Colour Label
0 -100 #FF0000 -100% loss of occurrence
100 0 #000000 No change
200 100 #00FF00 100% increase in occurrence
253 masked #FFFFFF Not water
Unable to calculate a value due to
254 -128 #888888 no homologous months
255 127 #CCCCCC No data
A palette for users with deuteranopia colour-blindness is also available on the Data Access section of
the website.
Seasonality
Purpose
The Seasonality map provides information concerning the intra-annual behaviour of water surfaces for a
single year (2021) and shows permanent and seasonal water and the number of months water was
present.
Description (from the paper)
A permanent water surface is underwater throughout the year, while a seasonal water surface is
underwater for less than 12 months of the year. In some places we do not have observations for all 12
months of the year (for example, because of the polar night in winter) and in these cases water is
considered to be seasonal if the number of months where water is present is less than the number of
months where valid observations were acquired. A second consideration is that some lakes freeze for
part of the year. However, during the frozen period water is still present under the ice layer, both for
lakes and the sea. In the methodology ice is treated as a non-valid observation, so the observation
period corresponds only to the unfrozen months. If water is present throughout the observation period
(that is, the unfrozen period), the lake is considered to be a permanent water surface. If the area of the
lake contracts during the unfrozen period, then the pixels along the borders of the lake no longer
represent water, and those pixels will be considered to represent seasonal water surface. Seasonality is
computed for every year. A single data set for the contemporary period (2021) is made available.
An example of the seasonality map is shown below for the Mopti area of Mali. The permanent water is
represented in dark blue and areas of seasonal water are shown in lighter blue.
Bands
The Seasonality dataset contains the following band:
Name Data type Description
The number of months that water was present from
seasonality unsigned int8
January 2021 to December 2021
Symbology
The Seasonality dataset has the following values and symbology. The values from 1 to 12 are discrete.
Value Symbol Colour Label
0 #FFFFFF Not water
1 #99D9EA 1 month of water
12 #0000AA 12 months of water (permanent water)
255 #CCCCCC No data
Recurrence
Purpose
The Recurrence map provides information concerning the inter-annual behavior of water surfaces and
captures the frequency with which water returns from year to year.
Description (from the paper)
Water recurrence is a measurement of the degree of inter-annual variability in the presence of water.
This describes how frequently water returned from one year to another (expressed as a percentage).
Recurrence refers specifically to the temporal behavior of water surfaces; unlike occurrence, recurrence
is not systematically computed over the full span of the archive, because water may not have been
present from the beginning to the end of the archive. Thus, we first have to define a ‘water period’—
that is, that part of the archive where water was present at least from time to time; the recurrence in
fact quantifies this ‘time to time’. The water period is established individually for each pixel. The water
period runs from the first month in the first year in which water is observed to the last month of the last
year in which water is observed of the entire 38-year period. In addition to defining the water period we
also need to define a ‘water season’ (not equivalent to a ‘wet season’). The water season is identified
from the monthly water recurrence and is defined as those months of the year that from time to time
have water. A ‘water year’ is a year with at least one water observation, while an ‘observation year’ is a
year with at least one valid observation within the water season. Water recurrence is then calculated as
the ratio of the number of water years to observation years. The count of the number of years starts
with the year in which water was first observed and ends with the most recent year in which water was
observed. Years that contain only observations outside the water season are not counted; we have no
way of knowing whether water might have occurred in the water season because we have no
observations.
An example of the Recurrence map is shown below for North-East Arkansas in the USA with areas of
water that are inundated regularly (whether seasonal or permanent) shown in blue and areas that are
flooded on an episodic basis shown in orange.
Bands
The Recurrence dataset contains the following band:
Name Data type Description
The frequency with which water returns from years to year
expressed as a percentage
recurrence unsigned int8
Symbology
The Recurrence dataset has the following values and symbology. The values from 1 to 100 are discrete.
Value Symbol Colour Label
0 #FFFFFF Not water
1 #FF7F27 1% recurrence
100 #99D9EA 100% recurrence
255 #CCCCCC No data
Transitions
Purpose
The Transitions map provides information on the change in seasonality between the first and last years
and captures changes between the three classes of not water, seasonal water and permanent water.
Description (from the paper)
The thematic maps and temporal profiles were used to identify a set of water classes that characterize
transitions between the first year in which representative observations were acquired and the last year
of observation. Representative years are identified by comparing each year in turn with the annual
pattern of monthly recurrence from the temporal profiles. These profiles identify months in which
water was observed, and indicate the percentage of valid observations classified as water in any given
month. A year is flagged as representative if it contains sufficient valid observations from any
combination of months to bring confidence to the determination of the presence or absence of water.
The overall level of confidence is determined by the annual sum of the monthly long-term recurrences
of observed months (per year). The rationale is that the likelihood of a real absence of water for a year
is higher if the water is absent for months showing a high long-term water recurrence than from one
showing small rates of recurrence. In the latter case the absence of water may be explained by a
seasonal shift, and does not confer enough confidence to conclude that water was not present later.
Therefore, we considered that if the sum of the recurrence of the observed months is greater than 100,
the absence of water observation brings enough confidence to consider that water was actually not
present. Conversely, a single water presence is enough to demonstrate water presence. The water class
in that representative year is then fixed as the ‘first’ year. The last year’s water class is always the class
assigned to the last year of observation (2021) because we have enough observations available within a
year during this period.
The following transitions were mapped: unchanging permanent water surfaces; new permanent water
surfaces (conversion of land into permanent water); lost permanent water surfaces (conversion of
permanent water into land); unchanging seasonal water surfaces; new seasonal water surfaces
(conversion of land into seasonal water); lost seasonal water surfaces (conversion of a seasonal water
into land); conversion of permanent water into seasonal water; and the conversion of seasonal water
into permanent water.
These conversions refer to changes in state from the beginning and end of the time series; they do not
describe what happened in the intervening years, so an unchanging water surface means that the
seasonality at that particular point was the same in the first and last year it was observed, and not
necessarily that it was stable throughout. Stability must be checked at the pixel scale by using the
longterm water history described by the temporal profiles plus the recurrence and occurrence maps.
There are instances where water is not present at the beginning or the end of the observation record
but is present in some of the intervening years. By tracking the inter-annual patterns of such
‘ephemeral’ events and their intra-annual characteristics, each such pixel can be classified as either
ephemeral permanent water (land replaced by permanent water that subsequently disappears) or
ephemeral seasonal water (land replaced by seasonal water that subsequently disappears), depending
on the majority of the observed seasonality during the period of water presence.
An example of the transitions map is shown below for the Mopti region of Mali.
Bands
The transitions dataset has the following band:
Name Data type Description
transition unsigned int8 The type of transition between the first and last year
Symbology
The Transitions dataset has the following values and symbology. The values from 1 to 10 are discrete.
Value Symbol Colour Label
0 #FFFFFF Not water
1 Permanent
#0000FF
2 New permanent
#22B14C
3 Lost permanent
#D1102D
4 Seasonal
#99D9EA
5 New seasonal
#B5E61D
6 Lost seasonal
#E6A1AA
7 Seasonal to permanent
#FF7F27
8 Permanent to seasonal
#FFC90E
9 Ephemeral permanent
#7F7F7F
10 Ephemeral seasonal
#C3C3C3
255 No data
#CCCCCC
Maximum water extent
Purpose/Description
The Maximum Water Extent provides information on all the locations ever detected as water over the
38-year period. It is the union of all of the other datasets. An example is shown below:
Bands
The maximum water exent dataset has the following band:
Name Data type Description
max_extent unsigned int8 A simple flag indicating if water was detected or not
Symbology
The Transitions dataset has the following values and symbology. The values from 1 to 10 are discrete.
Value Symbol Colour Label
0 #FFFFFF Not water
1 # 6666FF Water detected
255 #CCCCCC No data
Water History
To help understand how water is distributed temporally throughout the 1984-2021 period and
throughout the year, there are three datasets that capture the temporal distribution of water: the
monthly recurrence, yearly water history and the monthly water history. These histories are shown in
the Global Surface Water Explorer and users can click anywhere to see the full water history of a
particular place. The history datasets are currently accessible in Google Earth Engine but will be made
available for download in the near future.
An example of the three history datasets from the website are given below (from Extended Data Figure
6 in the paper).
a Examples of increasing surface water occurrence in Myanmar.
b Pixel based temporal profiles showing recurrence by month over 32 years (top), water history, by
seasonality class, by year over 32 years (middle) and monthly water presence for each year in the
water seasonality record, in this case 2007 (bottom). Collectively the graphs show that at this location
(Lat: 18.3928º, Long; 96.2633º) there are no observations in 1984-‘86, ’93, ’97 or ’98 (the gaps in the
middle graph), that prior to 2011 this was dry land, that the dam formed in 2011 and this point was
flooded sometime between April and September (bottom graph), but since then it has been
permanent water (centre graph), and that in the 32 years of observation water has not been detected
in June (no observations have been made in June since the dam filled
Monthly Recurrence
Purpose
The Monthly Recurrence dataset provides information on the intra-annual distribution of the water
and characterizes seasonality. Its main purpose is to see how water distribution varies over the year for
a particular place.
Description
The monthly recurrence is calculated in the same way as the annual occurrence but at the monthly
level. It is therefore a measure of how often water occurs between years for a particular month. There
are a total of 12 datasets for the monthly recurrence (one for each month) and each dataset has the
same bands and symbology.
The bands include a flag to indicate if the month in question has observations or not. Without this flag
it would not be clear from the data if there was no monthly recurrence whether this was because there
was no water detected or simply because there were no observations for that month. This may be the
case in high latitudes for example, because of the polar night in winter when there are no observations.
Bands
The monthly recurrence dataset is available in Google Earth Engine as an image collection with 12
images. Each image contains the following bands:
Name Data type Description
The recurrence value expressed as a percentage for
this month
monthly_recurrence signed int8
has_observations signed int8 A flag to indicate if the month has observations
Each image also has a metadata property to indicate the month of the year.
Yearly History
Purpose
The Yearly History provides information on the seasonality of the water over the 37-year period. It
contains the same information as the Seasonality dataset but for every year where observations are
available within the period 1984 to 2021.
Description
The description of a single year within the yearly seasonality classification is the same as for the
seasonality dataset.
Bands
The yearly history dataset is available in Google Earth Engine as an image collection with 38 images.
Each image contains the following band:
Name Data type Description
Classification of the seasonality of water throughout
the year.
waterClass unsigned int8
The values are as follows:
Value Description
0 No observations
1 Not water
2 Seasonal water
3 Permanent water
Monthly Water History
Purpose
The water detection history provides information on all the water detections at the monthly level
from March 1984 to December 2021. Its main purpose is to see when particular events occurred,
for example, to see when a dam was constructed in a particular place.
Description
The water detection history is the dataset from which many of the other datasets are derived (e.g. the
yearly seasonality classifications) and provides information on whether or not water was detected in a
particular month.
Bands
The monthly water history dataset is available in Google Earth Engine as an image collection with 454
images. Each image contains the following band:
Name Data type Description
water int Water detection for the month
The values are as follows:
Value Description
0 No observations
1 Not water
2 Water detected
Each image also has metadata properties to indicate the month and the year.
Metadata datasets
The Global Surface Water datasets also provide some ‘metadata’ information which are statistics on
the overall number of water detections, number of observations and valid observations that are
present in the 1984-2021 period. These statistics can be used as a proxy measure of confidence as the
precision of the metrics at any location improves as the number of valid observations increases.
Bands
The occurrence metadata dataset is available in Google Earth Engine and has the following bands:
Name Data type Description
detections signed int16 The total number of water detections
valid_obs signed int16 The total number of valid observations
total_obs signed int16 The total number of observations (i.e. scenes)
Using Symbology Files
To use the *.qml file in QGIS 2.16.3:
Double click on the layer to open the Layer properties dialog
At the bottom of the dialog, click Style | Load Style…
Navigate to the location of the *.qml file and click Open
To use the *.lyr file in ArcGIS 10.4:
Double click on the layer to open the Layer properties dialog
Click on the Symbology tab and change the rendered to Unique Values
Click ‘Yes’ when prompted to build an attribute table
Click on the Import Symbology button in the top right
Navigate to the location of the *.lyr file and click Add
Procedures for earlier versions of the software are similar. For more information see the online
documentation for QGIS here and for ArcGIS here.
Using the Metadata Files
The metadata files contain minimal information on the associated datasets. Much more information on
the method, derivation of data and technical specification is given in the Nature paper, this user guide
and the FAQ on the website.
To use the *.xml metadatafiles in QGIS 2.16.3:
Install the Metatools Plugin
Select the dataset you want to import the metadata into
Click Plugins | Metatools | Import Metadata
Select the *.xml metadata file
To use the *.xml metadatafiles in ArcGIS 10.4:
Open ArcCatalog and click on the dataset that you want to import metadata into
Click on the Description tab and then Import
In the Source Metadata box, enter the location of the *.xml metadata file
Set the Import Type as FROM_ARCGIS and click OK
Using Web Map Tiled Services (WMTS)
These mapping services provide the best option for simply displaying the Global Surface Water datasets
as they use the cached images that are used in the Global Surface Water Explorer. If you do not need to
analyse the water data but simply map it, then you should use the WMTS services as they will be the
fastest to load and use and you will not need to download any *.TIFF files. To use these services in
Desktop GIS:
In ArcGIS for Desktop:
In the ArcCatalog Window, click on GIS Servers and then double click on Add WMTS Servers
In the URL box, enter: [Link]
water/downloads_ancillary/WMTS_Global_Surface_WaterV2021.xml and click OK
Expand the 'Global Surface Water on [Link]' item and drag layers onto the map
In QGIS:
In the Manage Layers toolbar, click on Add WMS/WMTS Layer
Click New and enter a name (Global Surface Water) and URL
([Link]
water/downloads_ancillary/
WMTS_Global_Surface_WaterV2021.xml) Click
OK and click Connect - select a layer to add to the map
If you have any errors due to SSL security certificates then click Ignore and Save.
License
All data here is produced under the Copernicus Programme and is provided free of charge, without
restriction of use. For the full license information see the Copernicus Regulation.
Publications, models and data products that make use of these datasets must include proper
acknowledgement, including citing datasets and the journal article as in the following citation.
Citation
Jean-Francois Pekel, Andrew Cottam, Noel Gorelick, Alan S. Belward, High-resolution mapping of global
surface water and its long-term changes. Nature 540, 418-422 (2016). (doi:10.1038/nature20584)
If you are using the data as a layer in a published map, please include the following attribution text:
'Source: EC JRC/Google'
Contact
For more information and feedback please contact:
jrc-surfacewater@[Link]
