Automated fish kill species distribution and abundance acquired from drone imagery

Keywords: Coastal, disaster, mortality event, unoccupied aerial systems, UAS, UAV
Abstract Type: Poster Abstract

Abstract

Fish kills are large mortality events unrelated to an organism’s natural life cycle, predation, or harvesting that can significantly affect food webs, local economies, and fisheries. These events are increasing worldwide due to agricultural and industrial run-off and climate variations, but there is limited data documenting the species distribution, location, and abundance of dead organisms during these events. Unoccupied aerial systems (UAS, or drones) are ideal for monitoring fish kill events rapidly with limited field assistance, as they allow stakeholders to survey large areas from a safe distance. Moreover, dead organism distribution and abundance can be automatically detected and enumerated from UAS imagery using deep learning algorithms. To test the applicability of UASs for monitoring fish kill events, we surveyed seven shoreline sites using a DJI Phantom 4 RGB drone during a massive fish kill event in Tampa Bay, FL in 2021. Species-level dead organism ground-truthing data were collected via quadrat and float-by surveys at six locations. The total measured length of dead specimens included in quadrat surveys was also recorded. Preliminary results suggest that UAS imagery was an effective tool for monitoring dead organisms that have washed ashore but were not as reliable for monitoring dead organism abundance on the water. To automate organism enumeration, the YOLOv3 algorithm is being trained to detect organisms to morphology group, water cover, sand cover, and sargassum and seagrass washed ashore cover. Our goal is to share these methods with stakeholders interested in surveying areas impacted by fish kills at high spatiotemporal resolutions.