Geospatially enabled scientific workflows for reproducible environmental modeling and decision support

Despite the undoubted importance of scientific reproducibility, faithfully replicating an environmental modeling exercise remains, today, a challenge for modelers and geospatial scientists. Past experience has shown how a lack of transparency in models and the modeling process, combined with the inherent complexity and uncertainty of environmental data and models, can lead to a lack of reproducibility of environmental modeling. In turn, this can lead to a lack of warrantability in the outputs of environmental models and the decisions based on those outputs. Building on previous advances in reproducible research, this thesis aims to address this challenge, and to improve the reproducibility and warrantability of environmental modeling and associated decision-making processes. The approach uses scientific workflows to capture explicitly the entire process of environmental modeling, including all of the data and computational assets and steps needed to execute an environmental model. The hypothesis is that by increasing transparency, reducing uncertainty, and managing complexity, scientific workflows can contribute to demonstrated improvements in the reproducibility and warrantability of environmental modeling. The research explores the approach with reference to two major case studies, in solar radiation modeling and hydrological modeling. The case studies show how scientific workflows can a) aid in developing self-documenting environmental models with increased transparency and explainability; b) promote the integration of machine-learning support for model execution (for example, assisting in more transparent solar radiation model parameter-setting); and c) aid in decision support for model users (for example, in prioritizing new hydrological station locations). Ultimately, the research aims to illustrate how the application of geospatially-enabled scientific workflows to environmental modeling can contribute to wider understanding, uptake, scrutiny, and reliability of both environmental models and the outputs they generate. Such outcomes promise clear benefits for scientific research in geospatial sciences, and for applied evidence-based decision making more broadly.