|
Abstract:
|
Drone-based imagery is increasingly used to measure the size and condition of marine mammals, among other species. Drones fly above a target animal, and a picture is taken. The camera’s characteristics and altitude define a geometry problem that lets researchers compute the animal’s size based on how big it appears in the image. But, the animal’s size in the image and the altitude are observed with uncertainty. Uncertainty stems from image resolution and imperfect altimeters. Measurement errors can be estimated via a calibration study, where images are taken of references objects, whose exact length is known. We construct a hierarchical Bayesian model that uses calibration data to learn about measurement errors for several altimeters (i.e., laser-based and barometer-based), then yields posterior predictive distributions for the unknown measurements of the animals. The model’s hierarchical form lets us estimate relationships between lengths and widths of whales, which is a proxy for health. We also estimate uncertainty for length-based estimates of a whale’s maturity, and discuss extending the model to other animals, imaging problems, and measured quantities and relationships.
|
