The art and science of mapping: computing geological categories from field data

Like many activities in the geosciences, geological mapping of surface bedrock involves the construction of a model for a geographic region via field-based surveys. Individuals interpret field evidence to constrain possible histories and explanations, and these are regularly under-determined by available theory and data, resulting in multiple valid explanatory models where selection of the optimal model is often described as being an art as well as a science. This study empirically investigates this artistry by evaluating the correlation between data collected in the field and the geological map unit concepts interpreted from these data. Several geologists' data are selected from a completed field survey, and unsupervised and supervised categorization techniques provided by the self-organizing neural network are used to investigate the correlation between the selected data and interpreted concepts. Reported are results suggesting that the development of geological map unit concepts is influenced by theory, data, individuality and specific situations. Significant challenges in preparing largely qualitative data are also reported, and discussed are some broader implications related to the ability of computational techniques to capture and compute with the experiential knowledge of human agents in field-based situations. (C) 2004 Elsevier Ltd. All rights reserved.