Chemistry Across Multiple Phases (CAMP) version 1.0: An integrated\n multi-phase chemistry model

Matthew L. Dawson, Christian D. Guzmán, Jeffrey H. Curtis, Mario Méndez Acosta, Shupeng Zhu, Donald Dabdub, Andrew Conley, Matthew West, Nicole Riemer, Oriol Jorba

Introduction

Atmospheric chemistry modeling is a critical tool for understanding air quality and climate change. However, the field has long been hindered by "compartmentalization," where gas-phase chemistry, aerosol microphysics, and cloud processes are managed by separate, often hard-coded software modules. Chemistry Across Multiple Phases (CAMP) version 1.0 addresses this by providing an integrated framework that treats multi-phase chemistry as a single system, independent of the underlying physical representation of aerosols.

The Problem of Compartmentalization

Traditionally, chemical mechanisms are inextricably linked to a specific aerosol representation—such as "modal" for global models or "particle-resolved" for detailed local studies. This coupling means that moving a mechanism from one model to another requires extensive code rewriting and the use of specialized Fortran pre-processors like the Kinetic PreProcessor (KPP). This lack of flexibility slows down scientific progress and introduces potential errors when translating mechanisms between different scales.

The CAMP Framework

CAMP introduces an "abstracted" framework where the chemical solver does not need to know the specifics of how the host model tracks aerosols. Instead, it requests necessary physical parameters (e.g., mass, radius) through a standardized interface.

As shown in Figure 5, the traditional approach (a) involves separate models for different chemical phases, often requiring complex manual integration. In contrast, the CAMP architecture (b) uses a centralized "CAMP core" and an integrated chemical mechanism that communicates with the host model and solver via dedicated APIs. This design allows researchers to update complex multi-phase chemistry—including gas-phase reactions, aerosol partitioning, and emissions—at runtime without the need for code recompilation.

Methods and Implementation

The authors developed CAMP as a stand-alone C++ library built on three core pillars:

  1. Object-Oriented Abstraction: Using base classes for processes and parameters allows the model to treat gas and aerosol reactions with consistent logic.
  2. Abstracted Aerosol Interface: The AerosolPhase class ensures that chemical mechanisms are applied consistently whether the model uses bins (sectional), modes (modal), or individual computational particles.
  3. JSON Configuration: Chemical mechanisms are defined in human-readable JSON files, enabling runtime updates without recompilation.

Figure 6 illustrates this workflow. Users provide JSON files describing reactions of varying complexity. CAMP then maps these to "Modeled objects" which can be solved using various CPU or GPU solvers and applied to different aerosol representations within the host model. For solving stiff ordinary differential equations (ODEs), CAMP v1.0 utilizes the CVODE solver from the SUNDIALS package and the KLU sparse linear solver for efficiency.

Validation and Key Results

CAMP was integrated into two distinct host models for testing: PartMC (a stochastic, particle-resolved box model) and MONARCH (a 3D multiscale atmospheric chemistry model). Validation across these platforms yielded several key insights:

  • Cross-Representation Consistency: Box model tests showed nearly identical gas-phase ozone results across modal, sectional, and particle-resolved representations, proving the robustness of the decoupling.
  • Aerosol Dynamics: The study revealed that while chemistry remains consistent, the choice of aerosol representation significantly impacts physical outcomes. For instance, the modal approach reached equilibrium for Secondary Organic Aerosol (SOA) formation faster than the particle-resolved approach.
  • 3D Performance: In the MONARCH model, CAMP successfully simulated ozone and isoprene SOA over Europe and North Africa. Validation against European Environment Agency (EEA) station data showed a mean bias of less than 5 ppbv for most stations, proving the library is robust enough for large-scale regional modeling.

Future Directions

While CAMP v1.0 is a significant step forward, future work will focus on computational efficiency and physical complexity. Phase 2 of the project involves full GPU integration, which has already shown a 3x speedup in preliminary tests. Subsequent versions aim to include more complex physical processes such as deliquescence (salt dissolving into water), efflorescence (crystallization), and surface-specific reactions to further enhance the model's realism.

Conclusion

CAMP version 1.0 represents a paradigm shift in atmospheric modeling by democratizing model development. By separating chemistry from aerosol representation and moving to a runtime configuration, it allows researchers to focus on the science of atmospheric chemistry rather than the complexities of software engineering.

Summary updated

Abstract

A flexible treatment for gas- and aerosol-phase chemical processes has been\ndeveloped for models of diverse scale, from box models up to global models. At\nthe core of this novel framework is an "abstracted aerosol representation" that\nallows a given chemical mechanism to be solved in atmospheric models with\ndifferent aerosol representations (e.g., sectional, modal, or\nparticle-resolved). This is accomplished by treating aerosols as a collection\nof condensed phases that are implemented according to the aerosol\nrepresentation of the host model. The framework also allows multiple chemical\nprocesses (e.g., gas- and aerosol-phase chemical reactions, emissions,\ndeposition, photolysis, and mass-transfer) to be solved simultaneously as a\nsingle system. The flexibility of the model is achieved by (1) using an\nobject-oriented design that facilitates extensibility to new types of chemical\nprocesses and to new ways of representing aerosol systems; (2) runtime model\nconfiguration using JSON input files that permits making changes to any part of\nthe chemical mechanism without recompiling the model; this widely used,\nhuman-readable format allows entire gas- and aerosol-phase chemical mechanisms\nto be described with as much complexity as necessary; and (3) automated\ncomprehensive testing that ensures stability of the code as new functionality\nis introduced. Together, these design choices enable users to build a\ncustomized multiphase mechanism, without having to handle pre-processors,\nsolvers or compilers. This new treatment compiles as a stand-alone library and\nhas been deployed in the particle-resolved PartMC model and in the MONARCH\nchemical weather prediction system for use at regional and global scales.\nResults from the initial deployment will be discussed, along with future\nextension to more complex gas-aerosol systems, and the integration of GPU-based\nsolvers.\n
BibTeX
@misc{dawson2021chemistry,
  author = {Matthew L. Dawson and Christian D. Guzmán and Jeffrey H. Curtis and Mario Méndez Acosta and Shupeng Zhu and Donald Dabdub and Andrew Conley and Matthew West and Nicole Riemer and Oriol Jorba},
  title = {{Chemistry Across Multiple Phases (CAMP) version 1.0: An integrated\textbackslash\{\}n multi-phase chemistry model}},
  year = {2021},
  eprint = {2111.07436},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2111.07436},
}
Figure