Latent space atmospheric particulate science
Latent space atmospheric particulate science
Particulate matter air pollution is a major contributor to global illness and originates from a range of natural and anthropogenic sources. These particles can undergo agglomeration, whereby individual units cluster into larger, structurally complex assemblies. This process can span multiple spatial and temporal scales, from nanometre-sized particles interacting over seconds to minutes, to the formation of larger aggregates that form and evolve over hours or days. Understanding how particles agglomerate, and transport is essential for accurately predicting their dispersion, environmental impact, and potential health risks. However, traditional analytical models struggle to capture the multivariate, nonlinear nature of these processes. Recent advances in deep learning, in particular the use of latent space representations offer promising new tools for addressing this complexity. More specifically, latent spaces allow high-dimensional data, such as particle morphology and spatial-temporal measurements, to be projected into lower-dimensional manifolds where hidden structure and dynamics may become more interpretable. In this work, I explore how latent space methods can be applied to model particulate agglomeration and its evolution over space and time, providing an alternative framework for understanding aerosol behaviour beyond conventional techniques.
Grant-Jacob, James A.
c5d144d8-3c43-4195-8e80-edd96bfda91b
24 July 2025
Grant-Jacob, James A.
c5d144d8-3c43-4195-8e80-edd96bfda91b
Grant-Jacob, James A.
(2025)
Latent space atmospheric particulate science.
Artificial Intelligence for the Earth Systems.
(doi:10.1175/AIES-D-25-0042.1).
Abstract
Particulate matter air pollution is a major contributor to global illness and originates from a range of natural and anthropogenic sources. These particles can undergo agglomeration, whereby individual units cluster into larger, structurally complex assemblies. This process can span multiple spatial and temporal scales, from nanometre-sized particles interacting over seconds to minutes, to the formation of larger aggregates that form and evolve over hours or days. Understanding how particles agglomerate, and transport is essential for accurately predicting their dispersion, environmental impact, and potential health risks. However, traditional analytical models struggle to capture the multivariate, nonlinear nature of these processes. Recent advances in deep learning, in particular the use of latent space representations offer promising new tools for addressing this complexity. More specifically, latent spaces allow high-dimensional data, such as particle morphology and spatial-temporal measurements, to be projected into lower-dimensional manifolds where hidden structure and dynamics may become more interpretable. In this work, I explore how latent space methods can be applied to model particulate agglomeration and its evolution over space and time, providing an alternative framework for understanding aerosol behaviour beyond conventional techniques.
Text
AMS Latent Space JAGJ 29062025 Corrected_Accepted_Changes
- Accepted Manuscript
Restricted to Repository staff only
Available under License Other.
Request a copy
Text
aies-AIES-D-25-0042.1
- Version of Record
More information
e-pub ahead of print date: 24 July 2025
Published date: 24 July 2025
Identifiers
Local EPrints ID: 503476
URI: http://eprints.soton.ac.uk/id/eprint/503476
ISSN: 2769-7525
PURE UUID: ad431ff2-17fa-439f-a936-cd2d2a124699
Catalogue record
Date deposited: 04 Aug 2025 16:32
Last modified: 04 Sep 2025 02:12
Export record
Altmetrics
Contributors
Author:
James A. Grant-Jacob
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics