Machine-learned atmospheric chemical mechanismsEPA Grant Number: R840012
Title: Machine-learned atmospheric chemical mechanisms
Investigators: Tessum, Christopher , Riemer, Nicole
Institution: University of Illinois at Urbana-Champaign
EPA Project Officer: Chung, Serena
Project Period: August 1, 2020 through July 31, 2023
Project Amount: $399,469
RFA: Chemical Mechanisms to Address New Challenges in Air Quality Modeling (2019) RFA Text | Recipients Lists
Research Category: Air , Air Quality and Air Toxics , Early Career Awards
The objective of the work proposed here is to produce a machine-learned condensed chemical mechanism that provides accurate results in air quality modeling simulations under a wide variety of atmospheric conditions. The proposed work builds on preliminary studies where we have created a neural-network-based chemical mechanism that operates with fewer chemical species and orders-of-magnitude less computational cost than the reference mechanism it was trained on, while reproducing diurnal cycles with low error in 90% of randomly initialized simulations. The proposed project will focus on reducing error in the remaining 10% of cases.
We will 1) create a mass-conserving traditional chemical mechanism, which will allow us to enforce the neural network to conserve mass; 2) train an encoder-operator-decoder neural network to condense and emulate our reference mechanism with low error in 100% of randomly initialized simulations; and 3) implement the neural network chemical mechanism in the WRF-Chem atmospheric model and evaluate predictive performance.
We expect our project to produce a class of condensed chemical mechanisms that are computationally much faster than conventional condensed mechanisms—without reduced accuracy—and can be created from explicit mechanisms in an automated and flexible manner. This would serve to reduce the computational expense of air quality modeling for regulatory impact assessment and would potentially enhance the air quality modeling capabilities of stakeholders who may not have access to the computing resources required to run current-generation models, such as state, local, and tribal agencies.