HIVIS Hydro-Haze Dataset: Diffusion-Leveraged GAN Dehazing Driven by Classification

The USGS operates a nationwide network of ground-based cameras, where the collected images are stored in the cloud-based National Imagery Management System (https://api.waterdata.usgs.gov/docs/nims), and made available to the public through the Hydrologic Imagery Visualization and Information System (https://apps.usgs.gov/hivis). These cameras are colocated with USGS streamgage stations. As of January 2025, the network comprises approximately 830 cameras strategically distributed across major watersheds in the contiguous U.S., Alaska, the Caribbean, and the South Pacific regions. The cameras in Alaska are exposed to harsh weather conditions, which frequently degrade images, particularly due to haze. This degradation leads to data loss over multiple days at certain stations, especially those with a daily image upload frequency. In total, 6,587 images were collected from nine stations between October 1st, 2022, and October 20th, 2024. The dataset comprises 6,405 haze-free, 105 hazy, and 77 corrupted images.