Learning Coagulation Processes With Combinatorial Neural Networks. (9th December 2022)

• Record Type: Journal Article
• Authors: Wang, Justin L.; Curtis, Jeffrey H.; Riemer, Nicole; West, Matthew
• Is Part Of: Journal of advances in modeling earth systems. Volume 14:Number 12(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-09
• Subjects: coagulation -- aerosol -- machine learning; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2022MS003252
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: Simulating the evolution of a coagulating aerosol or cloud of droplets in a key problem in atmospheric science. We present a proof of concept for modeling coagulation processes using a novel combinatorial neural network (CombNN) architecture. Using two types of data from a high‐detail particle‐resolved aerosol simulation, we show that CombNN models outperform standard neural networks and are competitive in accuracy with traditional state‐of‐the‐art sectional models. These CombNN models could have application in learning coarse‐grained coagulation models for multi‐species aerosols and for learning coagulation models from observed size‐distribution data. Plain Language Summary: The climate impacts of aerosols and clouds strongly depend on the number concentration and size distribution of aerosol particles and cloud droplets. A key process in determining these size distributions is the process of coagulation, that is, the process of two particles or droplets colliding and forming a bigger particle. While numerical methods exist to simulate the evolution of particle size distributions, these come with considerable computational cost and furthermore require detailed knowledge of the physical process causing the collisions in the first place, which can contain large uncertainties. Our study demonstrates the development of a machine learning model that addresses both challenges using a new type of neural network. Key Points: We introduce a new combinatorial neural network architecture to coarse‐grain aerosol coagulation simulations Our model incorporates structural priors and conservation laws from the particle population balance equation Our network outperforms a fully‐connected neural network on the same task and is able to forecast aerosol states several hours in advance