“Informing Environmental Observing System Design with Machine Learning”

Abstract: Designing environmental observing systems involves deciding where and when to collect data to answer scientific questions and address management needs, often with limited resources and incomplete knowledge of the system. Drawing on work in physical oceanography and environmental science in Antarctica and the North American Great Lakes, this talk will explore how quantitative methods and machine learning can help inform these decisions. I will discuss approaches to characterizing uncertainty and assessing the value of additional observations, including recent work using DeepSensor, a framework based on convolutional Gaussian neural processes, to investigate sensor placement in the Great Lakes. Together, these examples illustrate how machine learning can help us make better use of environmental observations and design observing systems that are more effective at addressing scientific questions.

Speaker Bio: Dani Jones (they/she) is Research Faculty at the University of Michigan’s Cooperative Institute for Great Lakes Research (CIGLR). Their background spans physical oceanography, applied mathematics, and data science. Before joining the University of Michigan, they spent ten years as a physical oceanographer at the British Antarctic Survey, working on numerical modeling and observational data analysis. Their current research focuses on data-intensive approaches to Great Lakes science and management, including the use of machine learning to inform environmental observing-system design.