Data-Driven Observing Network Design for the Great Lakes and Southern Ocean

Overview

Environmental observations are expensive, sparse, and often collected in challenging environments. This raises a fundamental question: if we can only observe part of a system, where should we look?

This project explores how machine learning and statistical methods can help inform the design of environmental observing systems. My work focuses on understanding where new observations would provide the greatest information gain, how observing networks perform under different objectives, and how observations can be translated into improved environmental forecasts and scientific understanding.

Current applications span both the Great Lakes and the Southern Ocean, two regions where environmental variability, logistical constraints, and limited observational resources make observing system design particularly challenging.

Research Themes

• Uncertainty-aware machine learning for environmental monitoring
• Environmental sensor placement and network design
• Active learning and information gain
• Observing system evaluation and optimization
• Integration of observations, models, and forecasts
• Environmental decision-making under uncertainty

Current Projects

Great Lakes Observing System Design

In partnership with NOAA GLERL, GLOS, and regional stakeholders, I am developing machine learning approaches to evaluate candidate observing strategies for the Great Lakes. This work uses convolutional Gaussian neural processes and related methods to identify where additional observations may provide the greatest benefit for forecasting and environmental monitoring.

Southern Ocean Observing System Design

As Co-chair of the Southern Ocean Observing System (SOOS) Observing System Design Working Group, I contribute to international efforts to evaluate and improve observing strategies in one of the most data-sparse regions of the global ocean.

Why This Matters

Every observing system reflects choices about what information is most valuable and where limited resources should be invested. Machine learning can help us explore these choices more systematically, but it does not replace scientific expertise or stakeholder priorities.

The long-term goal of this work is to develop tools that help scientists, managers, and communities make better-informed decisions about environmental observations and monitoring.

Related Work

• Environmental sensor placement using convolutional Gaussian neural processes
• Great Lakes observing system design using uncertainty-aware machine learning
• Southern Ocean observing system design through SOOS