Behavioral characterization and Particle Filter localization to improve temporal resolution and accuracy while tracking acoustically tagged fishes

Abstract

Quantifying the fine-scale movement patterns and habitat use of active fishes has historically been challenging due to their scope of movement or the labor intensive nature of actively tracking potentially wide-ranging species. This project focuses on improving the localization accuracy and temporal resolution of an acoustically tagged fish by filtering the position measurements received from an acoustic receiver array. Using the k-means clustering algorithm, data sets are broken into groups which have similar fish speeds and yaw rates to yield a discrete number of movement behaviors characterized by the mean and standard deviation of speed and yaw. Next, a Particle Filter state estimator is proposed, in which position and speed state estimates of particles are used to calculate the most likely motion behavior, which in turn is used as a first order motion model to propagate the particle's fish state estimates forward in time. These predicted particle states are compared with the position measurements and then resampled as done with most Particle Filters. Offline processing of a shovelnose guitarfish (Rhinobatos productus) data set shows that the estimation of the fish's location is improved during periods of time when no measurements could be obtained when compared with two common filtering approaches.