High-resolution satellite mapping of Earth's surface water during the past 32 years reveals changes in the planet's water systems, including the influence of natural cycles and human activities. See Letter p.418
Everyone appreciates that the water cycle can vary, and can cause floods and droughts at its extremes. On page 418, Pekel et al.1 map the full range of this variability, as evidenced by our rivers, lakes and wetlands, using more than 3 million satellite images collected over the past 32 years. This globally consistent analysis documents both natural water variability and humanity's major influence on Earth's water systems, and will provide a valuable baseline for observations of the effects of future climate change.
Detailed maps describing the location and extent of rivers, lakes and wetlands are needed for many studies of Earth science, but the full global distribution and variability of these systems has not been clearly understood. Scientists have developed methods to map water bodies using satellite observations — for example, by detecting the characteristic reflectance of sunlight from water. But this is a particularly challenging task because the colour of water varies greatly depending on depth, the presence of suspended sediments and dissolved chemicals, and the angle at which sunlight hits the surface. In addition, some land surfaces (such as snow, ice and lava) have similar reflectance characteristics to those of water, which means that water-detection algorithms need to be developed and calibrated carefully.
The first global surface-water map2 made using satellite observations was developed in 2009, but computational power restricted the spatial resolution to 250 metres, which is too low to enable detailed mapping of smaller lakes and rivers. This was a problem, because statistical estimates3 suggest that millions of lakes less than 1 square kilometre in size could account for abput 40% of the global area of inland water. The situation has since improved: a global analysis of water bodies at 30-metre resolution was undertaken recently4,5 using images from the Landsat programme (the world's longest-running initiative for acquiring satellite images of Earth).
However, the location and extent of water bodies can change with time, in part because of natural processes such as flooding, sedimentation and channel migration, but also because of human processes such as dam construction and water abstraction. This creates a need for a global-scale, high-resolution analysis of information taken at different times — a complete map of surface-water dynamics. Such dynamics have recently been captured in maps that enable scientists to distinguish permanent rivers and lakes from seasonal water bodies such as flood plains6 and to explore the long-term trends of surface-water changes7, but these studies used only a subset of all the Landsat images available.
Pekel and colleagues' ambitious work uses the entire Landsat archive8 to map global surface waters — more than 3 million images collected between 1984 and 2015. To handle this petabyte-scale data set, the authors used Google Earth Engine (go.nature.com/2fdt80k), a freely available cloud-computing platform for analysing big data sets of satellite observations. The Landsat data set was produced using three satellites, and multiple operational issues affected the collection and quality of the data. This presented unique challenges, in addition to those associated with water's variable reflective properties. To overcome these challenges, Pekel et al. used a combination of expert systems (computer systems that use artificial intelligence) and visual analytics to identify the existence or absence of surface water for every pixel of Earth imagery, each representing a square of side 30 metres; this was done at monthly intervals over the 32-year period.
An understanding of the frequency with which water occurs at different locations is certainly a useful result of such an analysis. However, more-meaningful information and visualization of global-scale changes are required to cope with gaps in the data set that result from cloud cover and operational deficiencies, and to allow specific interpretation of different surface-water dynamics such as seasonal cycle and long-term trend. Pekel and colleagues therefore provide thematic maps depicting persistence (whether water is always present, or just sometimes), gains versus losses, the consistency of seasonal cycles, permanent versus seasonal water, and transitions between seasonal and permanent water during the period analysed (Fig. 1). The output of the analysis and the thematic maps are available through a user-friendly interface (go.nature.com/2gj81ap), allowing anyone to explore any location and understand what surface-water changes have occurred, without the need for complex analysis or massive computing power.
The authors' high-quality analyses and visualizations of the data reveal that there were 2.78 million km2 of permanent surface water and 0.81 million km2 of seasonal surface water on Earth in 2015. During the full period of the analysis, 162,000 km2 of permanent water were lost or became seasonal, whereas 184,000 km2 of new permanent waters were created at different locations. More than 70% of the losses were concentrated in just five countries (Kazakhstan, Uzbekistan, Iran, Iraq and Afghanistan) clustered in the Middle East and Asia, raising serious questions about water security and transboundary water management in that region. Most of the permanent-water gain correlates with reservoir construction worldwide, but the impact of climate change was also detected in lake expansion caused by melting glaciers in the Tibetan Plateau. Changes that occur across decades, such as those due to the recent drought in Australia, also stand out clearly.
Any analysis that quantifies surface water from historical data sets will have limitations. In this case, data gaps affect the accuracy of the seasonality information; resolution issues prevent analysis of small water bodies; vegetation obscures important wetlands; and the 16-day repeat cycle of Landsat observation means that events that occur on shorter timescales, such as floods, may be missed. These problems will be addressed in the future by using better optical and radar sensors and more satellites, and by integrating satellite-observed data into models of surface-water dynamics.
Despite the limitations, Pekel et al. have provided our best understanding yet of the changes in our planet's surface water. Their findings will be crucial to many Earth-science studies — such as climate-modelling efforts, or investigations of ecology at the interfaces between land and rivers — and for global water-management initiatives.