Neural nets used to differentiate a new species of Salpingoporella (Dasycladales, calcareous green algae) from the Urgonian limestone in southeast France
Marc André Conrad, Nicolaos Carras, Rajka Radoičić · Bollettino Della Societa Geologica Italiana · 2008
Artificial Neural Networks (ANNs), are used as an additional tool to differentiate Lower Cretaceous, morphologically closely related species of Salpingoporella. On each specimen, measurements are made on the inner and outer diameters of the thallus, the spacing of the verticils, and the number of laterals per verticil. «NeuralXL Classifier», a Microsoft Excel plug-in software, is used to calculate a user-defined number of data clusters (categories) for each test. Results are positive: in Salpingoporella, clustering dimensions with an ANN does help to make out a species, or lower rank taxon, especially when visual differences look insignificant. The quite rare Salpingoporella parapiriniae n. sp. is described from Barremian and Lower Aptian Urgonian deposits of southeast France. The protologue is based on regular criteria (essentially enumerative and dimensional values, dealt with by regression analysis) whose outcome is confirmed by the corresponding ANN study. The type-level of the new species contains an exceptional, almost complete section into the upper, fertile part of Korkyrella SOKAC ˇ & VELIC ˇ, 1981. Based on this section, Korkyrella may be a synonym of the genus Cylindroporella JOHNSON, 1954.