(Apologies for cross-posting)
Title: How Much is Too Much? Pitfalls and Opportunities in Data-Rich Ecological Research
Date/Time: July 21, Thursday, 11:00 AM – 12:00 PM EST.
Abstract:
Acoustic surveys are one of the most important tools for bat research, conservation, and development impact mitigation (e.g., mortality from wind turbines, etc.). Bats use echolocation to orient in their environment, meaning they emit ultrasonic calls and construct a "mental image" of their surroundings by the echoes their calls produce. Different bat species use different signal designs and so it is possible to identify species from recordings, but some species can have similar signal designs, and all species can alter their signals to some degree making this a task often left to experts.
In the past, bat detectors were expensive, sensitive, and limited in battery life, recording quality, and recording storage capabilities. Technological advances in the past 15 years have transformed the field, and today high-quality detectors that can record for months in the field and store hundreds of GBs of high-quality recordings are relatively inexpensive and user-friendly. The result is that the main bottleneck for analyzing this treasure of information is expert time and effort.
ML-based automated tools for analyzing acoustic surveys make it possible for non-experts and citizen-scientists to collect and analyze data on scales never before seen, enabling ambitious projects that provide knowledge of species distribution both on larger scales and in finer resolution, as well as information about activity levels and patterns. However, due to the similarity in signal design between several species, combined with the flexibility in signal design within species and even individuals, the results of automated tools suffer both from uncertainty - calls that couldn't be assigned to a species, and from identification errors - calls that were mistakenly assigned to the wrong species. The severity of these problems may differ depending on locality and season. One approach to this uncertainty is to accept it and acknowledge it when interpreting the results and deciding upon policies. The other is to audit the results manually, thus reintroducing some degree of reliance upon expert time and effort
In this talk, I'll describe some of the aspects of acoustic surveys, present the history, present, and a view for the future, and discuss the reasons for the difficulties of automating bioacoustics analysis.
Bio of the speaker:
Dr. Eran Amichai is an integrative biologist studying animal behavior and sensory biology to answer
questions in the field of ecology and evolution. He is a postdoctoral research associate at Dartmouth
College's EEES program where he studies sensory individuality in bats and seasonality of predator-prey interactions
between bats and katydids. Eran earned his MSc (2011) and PhD (2018) in zoology from Tel Aviv
University, during which time he was also a member of the team of scientists and conservation
practitioners that established Israel's national bat monitoring program and conservation policies.
https://eamichai.weebly.com/. Click or tap if you trust this link." data-linkindex="8">https://eamichai.weebly.com/
This webinar will be offered online via Zoom. Please register to receive the Zoom link prior to the webinar.
Registration link: https://libcal.dartmouth.edu/calendar/itc/2022DSAIW4. Click or tap if you trust this link." data-linkindex="9">2022 Summer Data Science and AI Webinar Series (Week 4)
------------------------------
Jianjun Hua
Statistical Consultant
Dartmouth College
------------------------------