× close
Pouya Hosseinzadeh (left), USU doctoral student in computer science, and faculty mentor Soukaina Filali-Bubrahimi (right), assistant professor in the Department of Computer Science, learn about machine learning to enhance water data collected by satellites. A description of the method was published in the AGU Journal. . He presented the research at USU’s 2024 Spring Finals Conference, which will be held March 26-27. Credit: Mary-Ann Muffoletto
Satellites surrounding the Earth collect a wealth of water data about the planet, but extracting useful information about oceans, lakes, rivers, and streams from these sources can be difficult.
“Water managers need accurate data for water resource management tasks such as monitoring lake coastal zones, detecting sea level rise boundary shifts, and monitoring erosion,” said Pouya Hosseinzadeh, a computer scientist at Utah State University. talk. “However, when reviewing data from currently deployed satellites, we face a trade-off between obtaining complementary data with either higher spatial resolution or higher temporal resolution. We’re trying to integrate data to provide that.”
Various data fusion approaches have limitations such as sensitivity to atmospheric turbulence and other climate factors that can cause noise, outliers, and missing data.
According to doctoral student Hosseinzadeh and his faculty member Soukaina Filali Boubrahimi, the proposed solution is a hydrological generative adversarial network known as Hydro-GAN. The scientists developed his Hydro-GAN model in collaboration with his USU colleagues Ashit Neema, Ayman Nassar, and Shah Muhammad Hamdi, and describe the tool in an online issue. water resources research.
Hydro-GAN is a new machine learning-based method that maps available low-resolution satellite data to high-resolution data, said Filali Bubrahimi, an assistant professor in the USU School of Computer Science.
“Our paper describes the integration of data collected by MODIS, a spectroradiometer aboard the Terra Earth Observation System satellite and the Landsat 8 satellite, both of which have varying spatial and temporal resolutions. I have,” she says. “We are trying to fill the gap by generating new data samples that improve the resolution of the shape of water boundaries from the images collected by these satellites.”
The dataset used in this study consists of image data collected over a seven-year period (2015-2021) for 20 reservoirs in the United States, Australia, Mexico, and other countries. The authors present the case of Lake Thar Taal, a saline lake in Iraq that is comparable in size to the Great Salt Lake and faces similar climate and exploitation pressures.
“Using seven years of data from MODIS and Landsat 8, we evaluated the proposed Hydro-GAN model for the contraction and expansion behavior of Thar Taal Lake,” Hosseinzadeh says. “Using Hydro-GAN, we were able to improve predictions about changing areas of the lake.”
Such information is critical to regional hydrologists and environmental scientists who need to monitor seasonal changes and make decisions about how to maintain lake water supplies, he says.
Scientists have demonstrated that Hydro-GAN can generate high-resolution data at historical time steps not otherwise available, in situations where large amounts of historical data are required for accurate predictions.
“We believe this will be a valuable tool for water managers, and as we move forward with similar models, we will be able to create data that includes information on topology, snow volume, streamflow, precipitation, and temperature in addition to imagery.” “This research will enable us to take a multimodal approach that provides “and other climatic variables” at USU’s 2024 Spring Finals Conference, March 26-27 in Logan, Utah. Mr. Hosseinzadeh, who made the announcement, said:
For more information:
Soukaina Filali Boubrahimi et al., MODIS-Landsat Water Spatiotemporal Data Augmentation Using Adversarial Networks, water resources research (2024). DOI: 10.1029/2023WR036342
