Modularizing Earth system models for interactive simulation

Earth System Models (ESMs) play a vital role in understanding and assessing climate change and other Earth system related issues. They are complex software systems, mainly programmed in Fortran, that undergo changes as science progresses. ESMs comprise different sub-models addressing specific aspects of the Earth system, such as ocean circulation. Their code is, partly, decades old. They can start as small models, which often evolve into large, complex models or are integrated into other models [15]. The ESMs are continually modified and enhanced to provide insights for specific research questions. This leads to numerous variants and versions for an ESM, which must be maintained as other scientists intend to base their research on new features and reproduce results. Thus, ESMs are long-living software and face typical risks, e.g., blurring module boundaries and architectural erosion, due to changes in functionality, hardware, and language features [7, 17]. The resulting architectural debts can limit further research and may harm the validity of scientific results. Furthermore, the lack of documentation and the loss of knowledge, as scientists move to other positions, limit the maintainability of ESMs. Our goal is to improve the understanding of the current architecture of ESM, modernizing and modularizing them to facilitate their quality and enable us to use them in interactive simulations.

Interactive exploration of Earth system simulations may have great potential to improve the scientific modeling process. It will allow monitoring of the state of the simulation via dashboards presenting real-time diagnostics within a digital twin world. Furthermore, it will be possible to attach assessment models, or any other software, to the Earth system simulation software without the need to intermingle their codes. Specific objectives include:

In the next section we start with a look at the state of the art for ESM and cross-domain data handling for ESM. Reverse engineering with static and dynamic analysis for modularizing ESMs is introduced in the following section. Employing digital twins for interactive simulation of ESMs is discussed, before we conclude the paper.

In Earth science, the primary research activities are Earth observation and Earth modeling, followed by data analysis and data comparison [14].

Essential to migrating an ESM from a monolithic state into an interactive simulation is an understanding of its architecture, sub-models, and interfaces between these sub-models, as well as suggestions for restructuring an ESM. Reconstructing the software architecture [8, 9] is then the basis for recommending restructuring options. Reverse engineering is the process of analyzing a subject system to create representations of the system at a higher level of abstraction [3]. Our first step for reverse engineering is to recover the software architecture based on static source code analysis and dynamic runtime observations. The source code analysis identifies procedure calls, direct accesses to shared global variables, and dataflow traces. The runtime monitoring allows one to observe and identify call-traces, is able to inspect library calls, and provides information on how often certain functionality is used.

Digital twins were created for industrial production and technology engineering processes, such as digital twins for ocean observation systems [1]. Their goal is to optimize design and operation of complex processes through a highly interconnected workflow, combining a digital replica of the process with real-time observations of the physical system. The Earth system digital twin combines simulations and near-real-time observations to monitor the evolution of the Earth system [2]. For each cycle, the simulation generates a background forecast data ensemble of the Earth system, which is compared to observation data throughout a time window. The promise of digital twins of the ocean, atmosphere, and cryosphere of the Earth amounts to providing different ways of interacting with one or more Earth system simulation/assimilation systems. Today’s Earth system simulations all rely on splitting the scientific process into a simulation/data production step and a subsequent data analysis process.

The promise of increased capabilities for inspection of a digital twin can be envisioned without giving up the split into data production and data analysis. However, achieving true intervention requires interaction of researchers with an Earth system simulation while it is running. Furthermore, many applications using the online output of an Earth simulation software, such as assessment models, would benefit from data at higher resolution than usually written to persistent storage, and hence the divide between data production and data analysis needs to be overcome.

We expect that interactive exploration of Earth system simulations and observations may have great potential to improve the scientific discovery process. Our approaches to address this include the following:

Based on the reverse engineering of ESM architectures, we intend to select bounded contexts, i.e., the foundation for decomposing the monolithic ESMs into modules when using Domain-Driven Design [5]. Such an approach was successful in the domain of information systems [16], and we intend to transfer this to the domain of ESM.

Our key goals are to answer the following research questions:

This twin-enabled interactive Earth system simulation software design aims at rethinking the flow of information within and into/out of Earth system simulation software by bringing together experts from the field of software engineering, ESM development, and the field of application development and user interface design.