Global Monitoring of Inland Surface Water Dynamics using Remote Sensing Data

Freshwater, which is only available in inland water bodies such as lakes, reservoirs, and rivers, is an important natural resource as it sustains every form of terrestrial life on Earth. However, freshwater is increasingly becoming scarce across the world due to a variety of natural and human reasons, which has resulted in the increased incidence of adverse events such as dwindling ground water, shrinking freshwater bodies, severe droughts, and devastating floods. This scarcity of freshwater not only poses a significant global threat to the sustainability of humans but also to the Earth's ecosystem. As a result, managing inland water has become one of the major 21st century challenges for the U.S. and the world. A global water monitoring system that can provide timely and accurate information about the available surface water stocks across the world at regular intervals of time is thus critical for managing the already scarce freshwater resources.

A global surface water monitoring system will also enable a number of advances in understanding the dynamics of water resources and its management. First, monitoring water dynamics can help in assessing the impact of human actions and climate change on the state of inland water bodies. As an example, the Aral Sea has been shrinking since the 1960s due to the undertaking of several irrigation projects by the Soviet Union, which has brought the lake to the verge of extinction in 2015. As another example, several glacial lakes in Tibet have experienced severe melting and subsequent expansion due to the rising temperatures and growing rates of urbanization and atmospheric pollution in the Tibetan plateau. A global monitoring system can help in assessing the impact of such human actions and climate change phenomena on the dynamics of water bodies. Second, information pertaining to the dynamics of inland water bodies will aid in discovering relationships between changes occurring in different water bodies and their interactions with other climatic processes, such as heat waves and precipitation extremes. Third, a global monitoring system would also facilitate the forecasting of water stocks and risks in the future, which when coupled with information about the projected water demands can help in devising policies for managing water in a timely and effective manner. This is especially critical since existing agencies for monitoring water bodies at local to regional scales do not freely share their information across national and international borders, even resulting in scenarios involving artificial scarcity of water due to incomplete and asymmetric information.

The potential in creating a global surface water monitoring system is enabled by the emergence of remote sensing data in the past few decades, acquired via satellites orbiting the Earth. Remote sensing data provides global coverage of a variety of physical attributes about the Earth's surface at fine spatial resolutions and frequent temporal intervals, which can be appropriately leveraged for distinguishing the surface of water bodies from land bodies. This offers a promise for learning predictive models that can automatically estimate whether a particular location on the Earth at a given time is water or land using remote sensing data. However, traditional predictive learning approaches are not well-suited for addressing the unique challenges faced in analyzing remote sensing data for global water monitoring, described as follows. First, a major challenge in learning predictive models for global water monitoring is the fact that land and water bodies appear very different in different regions of the Earth, due to the presence of varying geographies, topographies, and climatic conditions. Furthermore, the same land or water body can show different characteristics in remote sensing data at different times, due to the presence of Earth's seasonal cycles and inter-annual changes. This heterogeneity within land and water bodies makes it difficult to learn predictive models that differentiate between all varieties of land and water globally. Second, remote sensing data is plagued with a high level of noise and outliers, due to the presence of clouds, aerosols, etc., which can significantly impact the performance of traditional predictive learning approaches. These challenges have restricted the application of existing surface water monitoring approaches to local and regional scales, while no effort has yet been made to monitor the dynamics of inland surface water bodies at a global scale. This motivates the need for novel research in data-driven approaches for global monitoring of inland surface water dynamics.

We present a global surface water monitoring system that makes use of novel machine learning algorithms for handling a variety of challenges in using remote sensing data, e.g. the use of ensemble learning methods to account for the heterogeneity within land and water bodies, and the use of physics-guided elevation constraints to improve the prediction performance even in the presence of noise and missing values. In this talk, we will demonstrate some of the preliminary capabilities of our water monitoring system that is able to capture a variety of surface water dynamics, e.g. construction of new dams and reservoirs across the world, on-going droughts in Brazil and California, and changes in river morphology such as river meandering and delta erosion. For more information, please visit: http://z.umn.edu/monitoringwater.