EC Bayes Seminar Series: Laura Cartwright (Wollongong)

Joshua Bon

This is a virtual and live* event at QUT (GP). * COVID restrictions permitting.

We are delighted to have our third EC* Bayes Seminar presented by Laura Cartwright from the University of Wollongong.For virtual attendees: A Zoom link will be emailed prior to event.*Early Career: We want to highlight great research from early career researchers as opportunities for ECRs to present their work to broad audiences has declined.The physical meeting will take place in QUT Gardens Point O Block Room 520 (Room GP-O-520) - room subject to change.If you wish to attend in-person with morning tea provided please select this ticket option. Many thanks to our sponsors:QUT Centre for Data ScienceARC Centre of Excellence for Mathematical and Statistical Frontiers (ACEMS)Statistical Society of Australia (Queensland Branch)Title: Emulation of Lagrangian particle dispersion model sensitivities using a convolutional variational autoencoderAbstract:Flux inversion is the process by which sources and sinks of greenhouse gases are identified from observations of gas mole fraction. At the core of the flux inversion process is an atmospheric transport model, often a Lagrangian particle dispersion model (LPDM), that simulates particle trajectories in the atmosphere. These models are highly sophisticated, but are also highly computationally intensive. In related studies, researchers have proposed emulating the output of LPDMs at a small set of locations using a small subset of the LPDM inputs. However, existing approaches are limited to situations involving a single, known source of a specific atmospheric pollutant that could only have been emitted from that source. With greenhouse gases, the spatial sources and sinks constitute entire spatial fields. We thus need to emulate plumes over an entire spatial domain.In this talk I will present a method to emulate the sensitivity plumes of methane mole fractions to a methane flux field. We use a convolutional variational autoencoder (CVAE) to reduce the dimension of the sensitivities (via the encoder segment), and fit a spatio-temporal Gaussian Process emulator on this reduced-dimensional space. Using the decoder segment of the CVAE, we obtain corresponding plumes from emulated variables. We then use our model to emulate a collection of sensitivities obtained using the Numerical Atmospheric Modelling Environment (NAME). For comparison, we repeat the study using empirical orthogonal functions (EOFs) for dimensionality reduction. Based on performance metrics computed on the emulated sensitivities, the CVAE performs marginally better than the EOFs. We conclude that our emulation-based approach can be used to considerably reduce the amount of computing time needed to generate LPDM outputs for use in high-resolution flux inversions.

Our {communityGuidelinesLink} describe the sort of content we prohibit on Eventbrite. If you suspect an event may be in violation, you can report it to us so we can investigate.

To report other categories of prohibited or illegal content, submit {link}

EC Bayes Seminar Series: Laura Cartwright (Wollongong)

More events from Joshua Bon

Discover more events from Joshua Bon, from Science & Tech to other experiences you might love.

Still looking for the right event?

Explore all online events to browse and filter by date, category, and more.

EC Bayes Seminar Series: Laura Cartwright (Wollongong)

More events from Joshua Bon

Discover more events from Joshua Bon, from Science & Tech to other experiences you might love.

You might also like...

Browse more events with different dates, prices, and formats to find your next great experience.

Still looking for the right event?

Explore all online events to browse and filter by date, category, and more.