The use of remotely piloted vehicles in fisheries monitoring

Daniel Scott Auerbach

doi:10.7273/000005309

Population management of species currently relies on human observations to enumerate population status and trends. Field surveys are labor-intensive, often expensive to conduct, and may be invasive to wildlife and their habitats. In addition, surveys may be biased, resulting in the under or overcounting of individuals, impacting management decisions. To supplement traditional methods and increase survey extents, researchers have turned to aerial methods (airplanes or helicopters) for population surveys, yet these are dangerous, expensive, and sometimes biased. Remotely piloted vehicles (RPVs, drones) have been used as an alternative to data collection methods that enhance or supplement traditional population count methods. RPVs fill an observational gap between ground and satellite surveys in measuring and monitoring the environment. As RPVs and sensor technologies improve, researchers and natural resource managers need the basic scientific information to integrate RPVs into decision making across ecological settings. To achieve a successful integration into current methodologies, the use of RPVs must be evaluated in both accuracy and precision, compatibility with long-term datasets, and efficiency. This dissertation research develops and evaluates a set of RPV methods for counting salmon nests (termed redds) and compares their results against traditional ground counting methods, identifies factors influencing RPV-based counts, and tests the application of a deep learning model to increase efficiency. The goal is to provide a basic scientific understanding to improve RPV applications in aquatic and fisheries sciences. Chapter One developed an RPV method for counting salmon redds to estimate population size. This study was the first to provide strong empirical evidence of the potential of RPVs in population monitoring of salmon redds at very high-spatial resolution. Chapter 2 expands on the 2020 study and focused on reducing interobserver variability and identifying sources of sampling bias in RPV-based redd counts. Chapter 2 developed the necessary steps to integrate RPV-based redd counts into pre-existing long-term datasets. Chapter 3 applied a deep learning methodology for automated detection of redds from RPV imagery. This is the first deep learning model for counting redds from RPV imagery. Together these chapters provide the foundation on how to integrate RPV-based data into fisheries survey methods.

The use of remotely piloted vehicles in fisheries monitoring

Files and links (1)

Abstract

Metrics

Details