Geoscientific modeling and simulation helps to improve our understanding of the complex Earth system. During the modeling process, validation of the geoscientific model is an essential step. In validation, it is determined whether the model output shows sufficient agreement with observation data. Measures for this agreement are called goodness of fit. In the geosciences, analyzing the goodness of fit is challenging due to its manifold dependencies: 1) The goodness of fit depends on the model parameterization, whose precise values are not known. 2) The goodness of fit varies in space and time due to the spatio-temporal dimension of geoscientific models. 3) The significance of the goodness of fit is affected by resolution and preciseness of available observational data. 4) The correlation between goodness of fit and underlying modeled and observed values is ambiguous. In this paper, we introduce a visual analysis concept that targets these challenges in the validation of geoscientific models - specifically focusing on applications where observation data is sparse, unevenly distributed in space and time, and imprecise, which hinders a rigorous analytical approach. Our concept, developed in close cooperation with Earth system modelers, addresses the four challenges by four tailored visualization components. The tight linking of these components supports a twofold interactive drill-down in model parameter space and in the set of data samples, which facilitates the exploration of the numerous dependencies of the goodness of fit. We exemplify our visualization concept for geoscientific modeling of glacial isostatic adjustments in the last 100,000 years, validated against sea levels indicators - a prominent example for sparse and imprecise observation data. An initial use case and feedback from Earth system modelers indicate that our visualization concept is a valuable complement to the range of validation methods.
Originally published as:Dransch, D., Köthur, P., Schulte, S., Klemann, V., Dobslaw, H. (2010): Assessing the quality of geoscientific simulation models with visual analytics methods -a design study. -International Journal of Geographical Information Science, 24, 10, 1459-1479 DOI: 10.1080/13658816.2010 Assessing the quality of geoscientific simulation models with visual analytics methods -a design study Simulation models are essential means of scientific knowledge building and also the basis for decision-making. Because of their relevance, they have to be assessed thoroughly with respect to their quality. Simulation model assessment comprises two challenges: (a) modelers have to create a comprehensive mental image of the model's quality despite the massive multidimensional, multivariate, and often heterogeneous data; and (b) the model assessment process should be as efficient as possible. We face these challenges with a visual analytics approach. We aim at developing interactive visual representations which, in combination with present computational analysis methods, support the scientist's reasoning process to enhance the assessment of simulation models. In a design study, we analyzed two exemplary reasoning processes which cover the main model assessment procedures: the evaluation of the internal coherence of the model's structure and behavior and the assessment of its empirical validity. The analysis was conducted by means of a user-and task-centered approach which combines several knowledge elicitation techniques and task analysis concepts. We derived domain tasks as well as cognitive actions and developed and implemented interactive visualization components which supplement the statistical analysis methods already used. An informal qualitative user study shows that our visual analytics approach and tools help gain a more detailed mental image and hence a better understanding of the data and the underlying simulation model and allow for a faster and more comprehensive assessment of the simulation model.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.