Automated Integration of Large Geophysical Data Sets using Three Partitioning Cluster Algorithms: a Comparison
H. Paasche and D. Eberle
Event name: 11th SAGA Biennial Technical Meeting and Exhibition
Session: Numerical Methods
Publication date: 16 September 2009
Info: Extended abstract, PDF ( 860.76Kb )
Since the advent of modern desktop computers, attempts have been made in various geoscientific fields towards rapid, automated and objective information extraction from suites of co-located data sets. Multivariate unsupervised classification techniques, such as cluster algorithms, have been proven valuable tools for largely automated information extraction and are for example routinely used for structural exploration and integration of multi-spectral remote sensing data sets. However, so far very few attempts have been made towards using unsupervised classification techniques for rapid, automated and objective information extraction from large geophysical data sets. In this study, we employ the crisp k-means, fuzzy c-means (FCM) and Gustafson-Kessel (GK) cluster algorithms and compare their suitability for rapid and largely automated integration of complementary geophysical data sets comprising airborne radiometric and magnetic as well as ground-based gravity data. All three data sets cover a survey area of 5000 km2 located south-east of Johannesburg, South Africa. Integrated geophysical maps outlining dominant subsurface structures are obtained from each of the used cluster algorithm. Fuzzy cluster algorithms, such as the FCM and GK algorithm provide additional quantitative information about the trustworthiness of the detected subsurface units, which is considered very valuable when interpreting the finally obtained zonal maps. We will also show that the GK algorithm is most robust when it comes to the integration of data sets containing a few extreme anomalous values, e.g. as typically present in magnetic data sets, resulting in strongly skewed histograms of the data.