There are many unanswered questions regarding data in the natural sciences, especially those related to ecology and conservation. Daniel Sheldon and his team at the Machine Learning for Data Science (MLDS) lab at the University of Massachusetts Amherst College of Information and Computer Sciences’ (CICS) Center for Data Science (CDS) are working to answer those questions at the intersection of data science and conservation.
Within the lab, The Dark Ecology Project, a partnership with the Cornell Lab of Ornithology, has received significant attention. This effort uses over 20 years of radar data to develop novel computer methods for migration analysis and distinguish and track specific bird migration signatures for automated roost detection. Sheldon hopes to refine the accuracy of information that can be drawn from radar data by teaching computers to screen out errors like rain, which could be mistaken for a flock of migrating birds on radar.
Sheldon said the motivation behind this project came from his time working in Cornell’s lab before coming to UMass, and the distinct need to expand the progress of data analysis methods beyond just the domains of advertising, finance and insurance. “I believe strongly there are computational problems everywhere,” he said. “I think a lot of the work we have done in computer science, historically, has been applied to a relatively small slice of application areas. Now, so much of the problem is analyzing the data that comes from all of these domains.”
A majority of the projects in the MLDS lab are related to ecology. Sheldon is a lover of the outdoors and a person deeply concerned about its conservation, so for him, solving problems with data related to ecology, conservation and sustainability seemed obvious to Sheldon.
“There are huge pressing environmental concerns that we’d like to help solve,” he said, citing the influence of climate change on global migration as an example. “In ecology, we have very few data sets that can actually answer these questions. That’s why radar is so exciting because it is this super comprehensive data set about bird migration that goes back over 20 years, so we can really start to answer these questions.”
Sheldon was recently awarded a National Science Foundation (NSF) Faculty Early Career Development Award (CAREER), a continuing grant that will support research topics in machine learning, including applications in ecology and conservation, through 2023. This is one of three active NSF awards supporting Sheldon’s work. Sheldon said the funding will support his work with students on core machine learning computational problems and with collaborators to apply it to ecological problems. The award will help Sheldon’s efforts to better understand and correct systematic errors in citizen science data, something he worked on at Cornell.
On Dark Ecology, Sheldon collaborates with UMass CICS assistant professor Subhransu Maji, who “teaches computers to see.” Maji has designed computer vision techniques to recognize a large number of different types of objects in images. These ideas can be used to increase the value of citizen science data by automatically identifying the species of plant or animal in photographs taken by citizen scientists. For example, Maji’s algorithms have been applied to recognize hundreds different species of birds in photographs. Sheldon says, “In a lot of ways [citizen science data] is great because it’s this incredible boon of being able to do so much more than an individual scientist, but it’s not flawless,” he said.
In the lab, Sheldon is assisted by three Ph.D. students: Kevin Winner, Garrett Bernstein and Tao Sun. Winner, previously a visiting researcher at the Smithsonian Conservation Biology Institute (SCBI) on statistical methods for animal movement modeling using GPS collars, now focuses on statistical models for population ecology. He said he became interested in working with Sheldon at the MLDS lab because he grew up being taught to care deeply about the environment.
“I grew up near the Chesapeake Bay where conservation and ecology are relatively large facets of our education, and were always important topics to me,” he said. “When I found Dan’s lab and research and realized that it was possible to use machine learning for conservation and that there was this even broader community of people… I knew that was where I wanted to be.”
Committed to furthering the lab’s work for ecology, Sheldon said he will continue to expand existing and new projects to answer questions and solve problems in the domain. “I’m going to continue to push things across the spectrum. I’m interested in new applications,” he said. He envisages researching algorithms to help balance competing objectives between conservation and human development in river networks.
Sheldon said it is crucial that his work ultimately helps scientists in the field make decisions and inform sustainable development. “I want to think of ways the work can really have an impact,” he said.
Winner echoed this. “I love the ecological problems we’re tackling,” he said. “Seeing the impact our techniques are making on real-world problems has always been very fulfilling for me.”
Written by Morgan Hughes