Alouette van Hove: Guiding drones by information gain

Accurately estimating the location and strength of greenhouse gas sources remains a significant challenge in climate science, contributing to uncertainties in climate change projections. To address this issue, we develop a novel framework using atmospheric observations acquired via drones to infer the source location and emission rate of gas plumes. Through Bayesian inference, our framework aims to enhance the precision of greenhouse gas flux estimates. Ultimately, our objective is to facilitate the validation and calibration of existing models and metrics, thereby improving the reliability of climate change projections.

One aspect of our framework’s development is finding the optimal sampling path for drones to collect informative data in unfamiliar environments. In this presentation, we explore two sampling strategies in a synthetic study. Both strategies share a common goal: to maximize the information gained from sequential drone observations. We compare the nearsighted approach of infotaxis to a farsighted navigation strategy trained through deep reinforcement learning. Concluding the presentation, I will outline the subsequent steps in our research to implement the framework in real-world scenarios and discuss ongoing field experiments.

Alouette van Hove is a Ph.D. student at the Department of Geosciences of the University of Oslo. She has a background in environmental science, aerospace engineering and software development. Her research focuses on reinforcement learning and Bayesian inference to estimate greenhouse gas fluxes using drone observations. She is interested in computational methods and the application of machine learning techniques to address environmental problems.