Grantee Research Project Results
2023 Progress Report: Automated model reduction for atmospheric chemical mechanisms
EPA Grant Number: R840013Title: Automated model reduction for atmospheric chemical mechanisms
Investigators: McNeill, V. Faye , Fiore, Arlene M , Westervelt, Daniel , Henze, Daven
Institution: Columbia University in the City of New York
EPA Project Officer: Chung, Serena
Project Period: August 1, 2020 through July 31, 2023 (Extended to July 31, 2025)
Project Period Covered by this Report: August 1, 2022 through July 31,2023
Project Amount: $799,699
RFA: Chemical Mechanisms to Address New Challenges in Air Quality Modeling (2019) RFA Text | Recipients Lists
Research Category: Air , Air Quality and Air Toxics
Objective:
To develop an automated mechanism reduction algorithm for generating high quality reduced isoprene mechanisms.
Progress Summary:
Since the summer of 2022, our first paper on algorithmically reduced mechanisms was published, detailing the generation and performance of the AMORE-Isoprene v1 mechanism. This mechanism was generated using a combination of algorithmic and manual methods, and its performance was tested in CMAQ and in F0AM box models. Another paper is in review showing the performance of this mechanism in GEOS-Chem. Our results indicate that the AMORE-Isoprene v1 mechanism is highly accurate while saving significant run time in comparison to larger isoprene mechanisms. In our GEOS Chem testing, PM2.5, ozone, and formaldehyde were minimally affected by implementing our mechanism compared to a much larger baseline isoprene mechanism. The reduced isoprene mechanism led to 20-25% time savings over the baseline mechanism in the chemistry calculation runtime. Box model testing indicates that our mechanism is generally more accurate than existing highly reduced isoprene mechanisms of similar size.
We have been developing a fully automated mechanism reduction algorithm that we have tested on the isoprene mechanism and gecko-generated propane and camphene mechanisms. This new algorithm works by reducing mechanisms one species at a time. Species are quickly ranked in order from low to high yields from the source species, and species with the lowest yields are removed first. When a species is removed, any species that it produces is rerouted to the products of the species that produce the removed species, thereby minimizing the impact of the removal. This method allows species to be removed cleanly from a mechanism without significant error.
This reduction algorithm creates very high accuracy mechanisms up to 75-90% being removed. After that level of reduction, the accuracy is reduced, and we are now developing methods to improve accuracy for these smaller reductions. This new approach has multiple benefits. For one, it does not require any manual adjustment to the final output mechanism, meaning that the manual labor needed is greatly reduced. In addition, the reduction code is highly efficient, being able to reduce medium sized mechanisms (<1000 species) in under a minute, and larger gecko mechanisms (>100,000 species) within 15 minutes. Part of the efficiency is attributable to a novel, graph theory-based method of estimating species yields from a mechanism that does not rely on an ordinary differential equation (ODE) solver. This yield estimation method can reduce the time it takes to analyze a mechanism by an order of magnitude, greatly increasing the speed of the overall algorithm. This high computational efficiency allows for easy testing of multiple different input conditions and mechanism sizes. The algorithm runs completely in python, making it very accessible for any user.
At this stage, we are close to our goal of a fully automated reduction algorithm that can be applied to atmospheric oxidation mechanisms. There is still some work to be done to improve the algorithm and we intend to publish a paper on this algorithm in the first half of 2024.
We have also collaborated with another group to implement a genetic algorithm for the reduction of the isoprene mechanism. From this, we have pioneered a method to create accurate reduced mechanisms using an artificially intelligent approach. This method works by randomly generating isoprene mechanisms of a desired size, randomly changing them in small increments, and selecting the new mechanisms with the best performance. This methodology is meant to mimic evolutionary processes and is able to generate novel mechanisms for isoprene which perform competitively in our box model testing. We plan to publish our results for this method early in 2024.
Future Activities:
The project will continue on the same trajectory, employing various computational methods to develop algorithms for the reduction of chemical mechanisms. Within the next reporting period, a paper will be submitted for the newest AMORE algorithm for the automated reduction of generic chemical mechanisms that behave similarly to the isoprene mechanism. In addition, papers on genetic algorithms for mechanism reduction will be published as well.
We plan to produce several reduced mechanisms for useful organic species in the coming year which will be made available to the community for use in models. This includes camphene and alpha-pinene, and any other species of particular interest to the community.
Journal Articles on this Report : 1 Displayed | Download in RIS Format
Other project views: | All 4 publications | 2 publications in selected types | All 2 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Wiser F, Place BK, Sen S, Pye HOT, Yang B, Westervelt DM, Henze DK, Fiore AM, McNeill VF. AMORE-Isoprene v1.0: a new reduced mechanism for gas-phase isoprene oxidation. Geoscientific Model Development 2023;16:1801–21. doi:10.5194/gmd-16-1801-2023. |
R840013 (2023) |
Exit |
Supplemental Keywords:
Ispprene, mechanism reduction, atmospheric chemistry, secondary organic aerosol, graph theoryRelevant Websites:
MCNEILL GROUP @ COLUMBIA UNIVERSITY Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.