Accounting for Positional Uncertainty When Modeling Received Levels for Tagged Cetaceans Exposed to Sonar

sound propagation modeling methods and modeled positions of individual animals to estimate RLs in Exposure to anthropogenic sound can have a range four dimensions and to statistically describe uncerof negative behavioral and physical effects on tainty within volumes of water space where animals marine species and is of increasing ecological and were predicted to occur during exposure periods. regulatory concern. In particular, the response of By properly accounting for positional error in this marine mammals, and notably the family of crypstudy, it is clear that previous studies using single tic deep-diving beaked whales, to military sonar is median RL estimates drastically underestimate the a timely and complex issue. To make inference on full range of plausible values; ranges in estimated aspects of response by individual whales to noise of RLs here often exceeded 40 dB. We also demonany type, it is critical to either measure or systematistrate how ancillary data from visual focal follows cally estimate what received levels (RLs) the animal of tagged individuals can significantly narrow estiactually experienced. Various tools and techniques mated RL ranges. Further, we compared measured exist to monitor RLs and associated responses, each RLs on a calibrated acoustic tag to modeled RLs at with advantages and disadvantages. Most behavthe same position to evaluate our volumetric modioral response studies to date have used relatively eling results. While satellite tags record data over short-term (hours to a few days), high-resolution longer time frames, their substantial geospatial acoustic tags that provide direct RL measurements. error coupled with the unique deep-diving behavior Because of their short duration, these tags do not of beaked whales means that estimates of RL can allow for assessments of longer-duration basevary broadly and, consequently, that single point line behavior before and following a disturbance estimates from less robust approaches may be subthat may tell us more about the nature of response stantially in error. Accounting for this uncertainty within a broader context for tagged individuals. using robust statistical modeling is critical to fairly In contrast, longer-duration (weeks to months), characterize variance and effectively assess exposatellite-transmitting tags lack high-resolution kinesure-response relationships. matic data and the ability to directly measure RL. Herein, we address these issues and efforts to derive