USU Scientists Pioneer Hydro-GAN to Improve Satellite Water Data Analysis

USU Scientists Pioneer Hydro-GAN to Improve Satellite Water Data Analysis
by Clarence Oxford
Los Angeles CA (SPX) Mar 22, 2024
Utilizing satellites for gathering detailed water data across Earth's diverse aquatic landscapes presents significant challenges due to the varying resolutions of existing satellite imagery. Researchers at Utah State University have developed a novel solution to this problem, introducing a machine learning model known as Hydrological Generative Adversarial Network, or Hydro-GAN. This development enables the synthesis of high-resolution water data from lower-resolution satellite images, offering an innovative approach to water resource management.

Pouya Hosseinzadeh, a doctoral student at USU, emphasized the necessity of high-quality data for managing water resources, highlighting the difficulties faced when relying on current satellite technologies that provide data with either high spatial or temporal resolution but not both. Hydro-GAN aims to reconcile these differences, integrating satellite data to yield more precise information for applications such as coastal zone monitoring, sea level change detection, and erosion tracking.

The limitations of traditional data fusion methods, which are often affected by atmospheric disturbances and other environmental factors, lead to inaccuracies in the data. The Hydro-GAN model, developed in collaboration by Hosseinzadeh, his mentor Soukaina Filali Boubrahimi, and USU colleagues, offers a solution by generating accurate, high-resolution data samples from low-resolution satellite images.

Described in the team's publication in the American Geophysical Union journal Water Resources Research, Hydro-GAN leverages data from the MODIS instrument on the Terra satellite and the Landsat 8 satellite. Filali Boubrahimi, an assistant professor in USU's Department of Computer Science, explained that the model enhances the definition of water boundaries in satellite images, significantly improving water data accuracy.

The research, supported by the National Science Foundation, utilizes image data spanning seven years (2015-2021) from reservoirs across the globe. A case study on Lake Tharthar, Iraq, demonstrates Hydro-GAN's effectiveness in analyzing the lake's area fluctuations, offering valuable insights for regional hydrologists and environmental scientists.

Hosseinzadeh concluded that Hydro-GAN's ability to generate detailed historical data could greatly assist water managers in forecasting and decision-making processes. The potential for integrating this model with other data types, such as topology, snow depth, and climate variables, holds promise for enhancing global water resource management strategies.

Research Report:Spatiotemporal Data Augmentation of MODIS-Landsat Water Bodies using Adversarial Networks