We are developing an open-source computer vision algorithm to detect the location of dams from high resolution satellite imagery, enabling us to map the location of dams worldwide whose locations have not been systematically available. Knowing the locations of dams and reservoirs is a necessary first step for understanding human impacts to freshwater systems and their biodiversity, for effective management of the watersheds on which the reservoirs rely, and for projecting climate impacts to water supplies – all critical components of sustainably managing the water systems on which nearly every human relies. A partner on this project is Spring. Funding: National Geographic, Microsoft AI for Earth
Around the globe, millions of dams and reservoirs have been constructed to manage freshwater resources for human benefit. This infrastructure has had immense environmental impacts, both on freshwater biodiversity and on hydrological ecosystem services. Despite this, only the largest reservoirs — a small fraction of the total — have been mapped. Information on the location of dams of all sizes is vital in the face of increasing demands from existing and new dams, such as to meet growing irrigation needs for food production, and hydropower for renewable energy. Assessments based on currently available dam information, focused on larger dams, are therefore likely to greatly underestimate the magnitude of dam impacts and the importance of terrestrial ecosystems for providing hydrological services to people.
We combined high resolution remote sensing data with machine learning techniques to develop a generalized and replicable model for detection of smaller dams and reservoirs. By improving mapping of the location of small dams worldwide, we aim to contribute to efforts by researchers, non-governmental organizations, and governments to understand and mitigate the impacts of dams, to conserving and managing hydrological ecosystem services, and to sustainable development planning.