Advancing knowledge-rich intelligent systems in geology: Research on information integration, representation, and use

Advances in geoscience knowledge are contingent on the availability of intelligent systems that readily integrate and richly attribute multiple types and forms of data, that represent data and interpretations based on conceptualizations that are targeted to the problem, and that enable multi-scale, multi-attribute reasoning based on expert knowledge and practice. My research advances the knowledge and methods that underpin intelligent geoscience systems by addressing topics in integration, representation, and reasoning in geologic characterization research. My research has two main goals: 1) to characterize existing information use and decision-making practices that are used in geologic characterization investigations; and, 2) to contribute to the advancement of intelligent geoscience systems by extending foundational techniques in information integration and representation. In addressing these goals, the my research makes contributions to four topics within geologic characterization research: 1) the characterization of relationships between resource availability, information use, and model representations in 3-D geologic modeling; 2) the detailed characterization of information understanding and use by experts during geologic characterization investigations; 3) the development of a data integration framework that supports the multiple data streams and information behaviors used in geologic characterization research; and 4) the extension of existing research-documentation technology to deconstruct and describe information flow within a 3-D geologic modeling environment, in order to support the development of related knowledge engineering applications. Through a literature review, workflow analysis, and informal interviews with five geologic modelers, I deconstructed and analyzed digital 3-D geologic modeling workflows. I demonstrated that expertise availability has a critical role on differences in information use and model-outcome representation. The conceptualizations, representations, and visualizations required by modern digital 3-D geologic modeling are shown to be dependent on the software and hardware configuration of the underlying information system, and not simply dependent on the expertise of the modeling team. The collection of larger numbers and types of new observations (of targeted properties and at targeted locations) was found to occur with modeling efforts that developed more complicated conceptualizations and representations of geologic properties, but not necessarily more complicated model outcomes (i.e., visualizations). Results of a user study characterized how experts in geologic characterization understand their data and the information behaviors they use to interpret the distributions of geologic deposits. Using online interviews with 23 practicing geologists from the United States, Canada, England, and Denmark, the critical incident technique was used to ground semi-structured questions around how geologists understand and use multiple streams of data and the decision making strategies they apply to interpret the geology of their study areas. The analysis revealed that specific sets of information practices are used to reduce the uncertainties in both observations and interpretations. These techniques eliminate the perceived need to explicitly track uncertainty in most situations, reducing the cognitive effort that geologists exercise in any decision. In order to qualify the fitness of data for their characterization goals, geologists are shown to use multiple purpose-based assessments of their data. The use of data assessments and uncertainty reduction techniques underscore the importance that these experts place on assessing and building trustworthiness of their data and interpretations. In another key finding, unexpected pattern detection in data is shown to be the criterion that forced re-interpretation of data, often resulting in the identification and assignment of an alternative interpretation that better fits the identified pattern. Lastly, geologists were found to partition their expertise into experience and knowledge and to use each type for different reasoning purposes. Personal experiences were used to provide context for observed patterns and to support pattern learning and detection, while personal knowledge is used to recognize previously-learned patterns, to identify cues that helped reasoning with both unexpected and unrecognized patterns, and to apply patterns as constraints that enabled predictions across gaps in information. In another study, I developed the Reproducible Data Reuse (ReDaR) data integration framework – that supports data-supported geoscience reasoning while being compliant with FAIR data reuse guidelines, accommodating the multi-level assessment and building of trust, and supporting multiple data streams and observation-level reasoning. The design of the ReDaR framework is driven by a recognition of the semantic richness in datasets used for geologic characterization and an awareness of the potential difficulty of ensuring semantic alignment in datasets compiled from data streams that were collected and processed by different agents. Rich stream,- record-, and observation-level attribution is a guiding theme of the framework. One design priority was to include all attributes that might be used to establish data fitness, trustworthiness, or fidelity to the phenomena being observed, or to reason about the context and compatibility of observed values with specific geologic interpretations. Lastly, support for flexible, purpose-based assessments of data fitness or trust is a core priority of the ReDaR framework. Assessed values are treated as attributes to entities at the stream, record and data levels, and are designed to be determined based on customized reasoning involving other documented attributes. In a final study, I modified an existing research documentation tool, research process modeling (RPM), and used it to delineate process and artifact relationships within a 3-D geologic modeling environment. The changes made to the RPM method were designed to document different details about artifacts and processes – attributes that could better support knowledge-engineering applications. The results of the modified RPM process, applied to a case study 3-D geologic modeling workflow, revealed that the method was able to document important characteristics and relationships among the artifacts and processes that could be relevant to knowledge engineering applications. Further analysis of the RPM data showed that the documented information could also be useful for additional geoinformatics needs.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Informatics

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Keefer, Donald Allen

Contributors dc:contributor

• Blake, Catherine
• Ludaescher, Bertram
• Best, James L
• Phillips, Andrew C
• Wickett, Karen M

• Information Behavior
• Information Use Practices
• Personal Expertise
• Uncertainty
• Geologic Characterization
• Decision Making
• Integration
• Data Assessments

Statement dc:rights

• Copyright 2023, Donald Allen Keefer

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/121961

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F121961