The maps show different facets of surface water dynamics. Together the maps show where and when open water was present on the Earth's surface between March 1984 and October 2015. The meta-information in the web interface documents the number of valid observations at each pixel location, which provides users with a proxy measure of the degree of precision of all metrics provided by the website. Open water is any stretch of water open to the sky, and includes both freshwater and saltwater areas greater than 30mx30m. The maps display water surfaces that are visible from space, including natural (rivers, lakes, coastal margins and wetlands) and artificial water bodies (reservoirs formed by dams, flooded areas such as opencast mines and quarries, flood irrigation areas such as paddy fields, and water bodies created by hydro-engineering projects such as waterway and harbour construction).
Inland saline and freshwater lakes, rivers and wetlands collectively contain less than 0.01% of Earth's total water [most -96.5%- is held in the oceans and seas, with the remainder held by ice caps, glaciers, ice and snow, groundwater, held in the soil, contained in biological cells (including us!) and in the atmosphere]. But surface water is the most accessible and affects many aspects of our world. It affects the exchange of heat, gas and water vapour between the planet's surface and atmosphere. Water is the engine behind the distribution, movement and migration of Earth's plant and animal life and is just as essential for humans. It affects our capacity to grow crops and manage animal grazing lands, to run our industrial processes, to manufacture goods, it influences the movement of disease-vectors, toxins and pollutants, it generates energy directly (hydroelectric) and indirectly (thermoelectric), it is an essential part of our transport network, and forms part of our recreational, cultural and sporting world.
The maps were created from individual full-resolution 185 km2 global reference system II scenes (images) acquired by the Landsat 5, 7 and 8 satellites.
The images, distributed by the United States Geological Survey (USGS), were processed to Standard Terrain Correction level (Level 1T). The Landsat 5 Thematic Mapper (TM), Landsat 7 Enhanced Thematic Mapper plus (ETM+) and Landsat 8 Operational Land Imager (OLI) acquire multispectral imagery at 30m resolution in six visible, near and shortwave infrared channels, plus thermal imagery at 60m (TM and ETM+) and 100 m (OLI). 1,823 TB of data were used (3,066,102 individual images). Each satellite is in a near polar orbit, and provides global coverage every 16 days. The individual satellite orbits are such that when two operate concurrently there is an eight-day revisit period. Landsat 5 was launched on 1 March 1984 and collected TM imagery until November 2011. Landsat 7 was launched on 15 April 1999, and its ETM+ is still imaging (although 22% of each scene is missing since 31 May 2003, when the scan line corrector failed). Landsat 8 was launched on 11 February 2013 and began operational imaging April 2013. It too is imaging today.
The spectral reflectance characteristics of different land-cover types diverge at certain wavelengths. We used thermal imagery and the contrasting spectral properties of water and other features (including snow, clouds, shadows, bare rock and vegetated land) in the Landsat sensors' six visible, near and shortwave infrared channels to separate pixels acquired over open water from those acquired over other surfaces. The pixels were classified using an expert system. This is described in this paper (Pekel et al. 2016).
The spatial and temporal variations in the presence of surface water are captured in a single product called Surface Water Occurrence. This gives the frequency of occurrence of water on the land surface over a given time period. The map covers 32 years (from March 1984 to October 2015). The map records water occurrence in monthly time-steps. Some locations are underwater throughout the period of observation (100% of all observations are classified as water), others are underwater for a few months of every year (often for the same months each year), some are only underwater on an episodic basis, and some have never been underwater (0% classified as water).
Two separate 15-year periods (from 1984 to 1999, and 2000 to 2015) are used to build a map that shows the change in the intensity of surface water occurrence (Surface Water Occurrence Change Intensity). The change in water occurrence intensity between the two periods is derived from homologous pairs of months (i.e. same months containing valid observations in both periods). The difference in the occurrence of surface water was calculated for each homologous pair of months. The average of all of these differences constitutes the Surface Water Occurrence Change Intensity. The scores are averaged to account for variations in data distribution over time (i.e. seasonal variations and frequency of valid observations) and to provide a consistent measurement of the occurrence of change.
The resulting map shows where surface water occurrence increased, decreased or remained across the 32 year periods. Both the direction (i.e. increase or decrease) and intensity of the changes are documented. Areas where water has been detected with equal occurrence (considering a tolerance of +/-15%) in both periods are shown in black on the map (i.e. water presence, with no change). Locations where surface water occurrence decreased across the 32 years period are mapped in red and locations where the water occurrence increased are mapped in green; in both cases, brighter tones indicate greater changes in intensity. Grey areas represent locations where there are no pairs of homologous months between the two periods, and for which we cannot compute meaningful measurements of change.
Surface Water Seasonality describes the intra-annual distribution of surface water. It discriminates between 'permanent' and 'seasonal' water surfaces for any given year. A permanent water surface is underwater throughout the entire year, whilst a seasonal water surface is underwater for less than 12 months of the year. In some places, we don't have observations for all 12 months of the year (e.g. because of the polar night in winter). In such cases, water is considered to be seasonal if the number of months during which water is present is less than the number of months for which valid observations were acquired (i.e. water has been detected in some but not all months during which observations were possible).
Water Recurrence is a measurement of the degree of variability in the presence of water from year to year. It describes the frequency with which water returned to a particular location from one year to another, and is expressed as a percentage.
The "Water Transitions" map displays the changes in surface water classes (no water, seasonal, permanent) between the first and last years in which reliable observations were obtained (see Pekel et al. 2016 for details). The Water Transition map documents the following:
The complete history of any water surface can be accessed at the pixel scale as temporal profile charts that present the water recurrence and observation records for a particular period of time. These profiles allow users to identify specific months or years during which conditions changed, e.g. the date on which a new dam was created, or the month or year in which a lake disappeared.
Three histograms are provided. The monthly recurrence profile shows the distribution and seasonality of water during the year, and provides information on the water recurrence for each month. The bar chart shows the percentage recurrence of water (on the y-axis) for each of the 12 months of the year (along the x-axis). The water history chart shows the classification of the surface water (no water, seasonal or permanent) for each year in which valid observations were acquired, and can also display the presence and observations of water by month within any single year.
The accuracy of the water map was calculated using over 40,000 control points from around the world and across the 32 years. The full validation protocol and results are provided in Pekel et al. 2016. In summary the validation results show that the water detection expert system produced less than 1% of false water detections, and that less than 5% of water surfaces were missed.
The estimations provided are statistically robust, as they are derived from the analysis of over three million images collected over 32 years, and which have been individually processed using a high-quality classification algorithm.
As the precision of the estimates improves with greater numbers of valid observations, the number of valid observations available for each single pixel can be used as a proxy of the degree of precision (e.g. the estimation of metrics derived from 600 images are much more precise than those derived from 30 images).
Given that the number of valid observations collected over the 32 years varies from region to region, according the observation conditions (e.g. cloud cover) and acquisition plan, the degree of precision is different from one region to another.
The meta-information layers document each single 30mx30m location at the global scale. Using the web interface, the numbers of valid observations and areas classified as 'water' can be displayed by clicking on the map. These details are provided for each month in the time series, and can be used as a proxy of the degree of precision.
Three histograms are generated. First, monthly recurrence shows the intra-annual distribution of the water, and characterizes water seasonality. It also provides information on the water recurrence for each single month. Second, a water history chart shows the class (land, seasonal water and permanent water) for each year in which valid observations were acquired. Third, month-by-month presence of water and observations within any single year can be extracted.
This dataset is released with a full and open access conforming to the Copernicus Regulation and can be accessed from the download section.
Publications, models and data products that make use of this datasets must include proper acknowledgement, including citing datasets and the journal article as following:
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)
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)