Comments: Paper by Lacassie et al. (JGE, 91, 81–98 (2006))

Reply: Comments by Colliston and Schoch (2008) on paper by Lacassie et al. (JGE, 91, 81–98 (2006))

“…In particular, the subtle geochemical associations found by Lacassie et al (2003Lacassie et al ( , 2004, using GCS maps, indicate the presence of an unexpected geochemical signature, not seen by previous studies using more conventional methods, which were later confirmed by Ortiz and Roser (2006). Therefore, contrary to Colliston and Schoch (2008), we are confident that the obvious consistency of neural network clusters with lithostratigraphic position of samples collected at various localities throughout the Aggeneys-Gamsberg District is an important contribution to the understanding of the Bushmanland Group. Such consistency should be entirely unexpected, if one assumes the schist units were unrelated to one another, as repeatedly proposed by Colliston and Schoch (1998, 2002, 2008.…”

“…Furthermore, we would like to stress that our analysis is based on a subset of whole-rock geochemical data for the three most important schist units of the Bushmanland Group in the Aggeneys-Gamsberg District, presented by Lipson (1990;see Lacassie et al, 2006, p. 87). As such, we did not produce any new analytical data as erroneously invoked by Colliston and Schoch (2008).…”

“…We appreciate the comments by Colliston and Schoch (2008) on our recent publication (Lacassie et al, 2006) that highlights the importance of understanding the structural complexity of the Bushmanland Group. While we agree with the authors that the structural complexities require a vigilant approach to any stratigraphic analyses, we disagree that the structural and stratigraphic complexities prohibit stratigraphic correlation within the Aggeneys-Gamsberg District or regionally (Bushmanland Subprovince).…”

“…Colliston and Schoch (2008) are regrettably mistaken in assuming that samples were "grouped on lithological basis." Indeed we clearly state that "a priori knowledge was not used during the unsupervised training of the GCS map" (emphasis added, Lacassie et al, 2006, p. 90).…”

“…Based on the aforementioned, the clustering or group separation of the data in the trained network is a natural characteristic of the samples, irrespective of their lithological nature or lithostratigraphic position, and indicates that the geochemical associations are not merely "geochemical correlations between rocks with similar lithology" as suggested by Colliston and Schoch (2008). Furthermore, the results of Lacassie et al (2003Lacassie et al ( , 2004 show that the GCS analysis is sensitive to subtle geochemical differences between stream sediments of the same fluvial system located at relatively close geographical positions, i.e.…”