Abstract
Increasing computing power and the growing role of simulation in Earth systems science have led to an
increase in the number and complexity of processes in modern simulators. We present a multiphysics
framework that specifies interfaces for coupled processes and automates weak and strong coupling
strategies to manage this complexity. Process management is enabled by viewing the system of equations
as a tree, where individual equations are associated with leaf nodes and coupling strategies with
internal nodes. A dynamically generated dependency graph connects a variable to its dependencies,
streamlining and automating model evaluation, easing model development, and ensuring models are
modular and flexible. Additionally, the dependency graph is used to ensure that data requirements are
consistent between all processes in a given simulation. Here we discuss the design and implementation
of these concepts within the Arcos framework, and demonstrate their use for verification testing and
hypothesis evaluation in numerical experiments.
