Predicting the Time Evolution of Dispersing Atmospheric Clouds using Deep Neural Networks
Abstract
A better understanding of how gas clouds move through complex environments is important for many applications, including tracking greenhouse gas emissions, air pollution, and atmospheric transport and dispersion. Computer models can be used to simulate the time evolution of gas clouds, but they can be computationally expensive. Alternatively, data-driven methods may provide a faster way of predicting gas cloud evolution. Here, we present results demonstrating the feasibility of using deep neural networks to predict the time evolution of a gas cloud moving through an urban area. Using variations of spatial-temporal long short-term memory (ST-LSTM) neural networks trained on samples of the time behavior of gas clouds, we successfully predict the late time dynamics of the clouds as they dissipate across an urban area when given only early time inputs, even in situations where the clouds are bifurcated by obstacles. Our presentation summarizes the gas cloud data, ST-LSTM models, and evaluation metrics used in this study.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMGC45F0872L