Title: Tracking the Expansion of Small Reservoirs in Brazil (1984–2025) via Deep Learning

Abstract:

In Brazil, water research has largely neglected the extensive damming of small waterways driven by agricultural needs, including livestock watering, small-scale hydropower, irrigation, and aquaculture. These pervasive dams and their resulting reservoirs significantly impact water temperature, stream connectivity, aquatic ecosystems, greenhouse gas emissions, and evaporative losses. However, mapping these structures remains difficult due to the challenge of reliably identifying small water bodies and differentiating man-made reservoirs from natural lakes, leading to their exclusion from most regional and global datasets.

To bridge this gap, we developed a deep learning computer vision model capable of accurately segmenting stream-fed surface water reservoirs smaller than 1 km² in Brazil. By utilizing data from Landsat satellites 5 through 9, we applied this model across a 41-year period (1984–2025) to generate annual maps for the entire nation, allowing us to assess temporal changes in reservoir count, size, and distribution. Our analysis reveals that the number of identified reservoirs increased nearly fourfold, rising from 263,913 to 996,245. Concurrently, the total surface area expanded from 3,510 km² to 8,550 km².

To our knowledge, this study presents the first country-wide annual dataset documenting the four-decade evolution of small reservoirs. The resulting public maps underscore the magnitude and cumulative effects of small stream impoundments throughout Brazil, offering critical insights for the sustainable management of freshwater ecosystems and water resources.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC