2005 Salt Lake City Annual Meeting (October 16–19, 2005)

Paper No. 2
Presentation Time: 8:15 AM


FANG, YC, Paleoclimatology Group, Byrd Polar Research Center, 1090 Carmack Rd, 283 Scott Hall, Columbus, OH 43210, SCHWARTZ, Frank W., Geological Sciences, Ohio State Univ - Columbus, 125 S Oval Mall, Columbus, OH 43210-1308 and SCHINCARIOL, Robert A., Earth Sciences, Univ of Western Ontario, Biological and Geological Sciences Building, London, ON N6A 5B7, Canada, fang.29@osu.edu

One of the great challenges in ground-water modeling is the lack of information about some site of interest. Especially important in this respect are hydraulic conductivity data, which provides a key control on ground-water flow. Therefore, methodologies that contribute to understanding of the distribution of hydraulic conductivities in a porous medium are have potential to contribute to the improvement of models. Continuing experience suggests that traditional practice of making direct measurements in a few wells or permeameter measurements is inadequate, producing models that are unsatisfying sometimes contribute fundamental data errors. This study aims to develop new field-based approaches capable of estimating patterns of hydraulic conductivity. To this end, image processing and modern data mining techniques are being developed to predict hydraulic conductivity values from digital images of outcrops and eventually borehole walls. Schincariol and co-workers collected the digital images from a large outcrop of sand and gravel in a pit near London, Ontario, Canada. At selected locations on the outcrop, cores provide hydraulic conductivity measurements. Our basic approach involves using the very large database represented by the photograph and processing through data mining to extract information. In other words, data mining involves the extraction of interesting knowledge, such as rules, regularities, patterns, constrains, and etc. from data in large datasets. Among the powerful tools of data mining, pattern recognition and classification are being developed to better understand the architecture of sand and gravel deposits. Self-Organizing Maps (SOMs) is one of the most useful clustering tools and is applied in this study to cluster small images extracted from the outcrop along with 122 sampling points to predict the hydraulic conductivities for the whole section of the outcrop. The outcome of this analysis is a detailed, two-dimensional map of the hydraulic conductivity distribution.