Reasoning and decision making in geology typically require the integration of several different datasets. Each data set will have multiple attributes containing some level of uncertainty (i.e., inaccuracy, ambiguity, or incompleteness). Ideally, uncertainty in attribute values will be based on what is known about the sampled object and the reported observation values, and will be represented by a distribution of possible values. In a database with this design, this knowledge about the sampled object and the distribution of possible values would be used as constraints to queries that enable robust reasoning about the system. Because designing a database to use this type of knowledge is difficult, data are typically represented using only one value per attribute. However, representing an uncertain attribute with only one value will reduce the accuracy and robustness of resulting interpretations because all knowledge about the uncertainty in attribute values is lost. We have designed and developed a geologic database that represents imperfect attribute values within several different glacial sediment data streams (i.e., continuous core, split spoon samples, wireline logs, ERT profiles), and that supports reasoning about lithologic and stratigraphic picks from these data. The domains for the uncertain attributes include nominal categories (e.g., sand and gravel, shale), ordinal categories (e.g., well sorted, late Pleistocene), and numeric values. The database design also allows for identification and management of intentionally or accidentally incomplete (i.e., missing) values. We will discuss a conceptual model and representational strategy for managing multiple uncertain attributes within an integrated geologic database of multiple data types. We will discuss how this design can align reasoning and analysis of lithologic and stratigraphic picks, and why a geologic database that captures uncertainty is the first step to more effectively exploring the space of valid interpretations within any form of geologic reasoning (e.g., stratigraphic log, geologic map, 3-D framework model). Lastly, we will discuss why additional technologies may be needed to implement effective queries about sediment distribution, to maintain better fidelity to the uncertainty in both the observations and interpretations of geologic data, and to implement an more robust uncertainty-aware geologic knowledge-based system.
|Published - 2021