Automating Multivariable Workflow Composition for Model-to-Model Integration

Many societal-relevant challenges, including environmental ones, require comprehensive approaches that integrate decoupled data, models, and perspectives. Integrating data and models is critical for these approaches but can also become cumbersome. Computational workflows are widely used to integrate heterogeneous data and computational processes within or across domains. However, creating computational workflows may require computational and domain expertise not necessarily possessed by potential users. This paper presents our efforts to enable automated multivariable workflow composition implemented as a workflow composer in the Sustainable Water for Integrated Modeling (SWIM) platform. We describe the uninformed search algorithm used in the workflow composer and an initial evaluation with a case study that requires integrating two water (balance) models that cover the Middle Rio Grande in the U.S. Southwest region. Preliminary results show that the evaluation of integrating models should not only consider the technical and scientific perspective but also how users understand and use the results of these complex systems. Efforts toward automating the model-to-model integration can significantly support scientific endeavors and decision-making by enabling various stakeholders to use scientific models.