A Neural Network for Spectral Analysis of Stratigraphic Records

Astronomically controlled variations in the climate induce cyclic trends in the sedimentary process if being effectively registered by the depositional process, are easy to detect in shallow marine carbonate rocks. One of the main difficulties to be solved in order to choose among the registered periodicities is the conversion from the spatial (i.e. recurrent variations along the strata sequence) to the temporal domains of the astronomically induced frequencies present in the rock record. We discuss here how this problem can be circumvented by teaching a neural net to recognize armonics in the signal. The application to two sequences in the Cretaceous of Southern Italy has shown this approach to be particularly effective, and has confirmed the existence of Milankovic-type periodicities in the examined records, where climate, sediments and biota concomitantly react to the variation of the solar constant induced by secular perturbations of the Earth's orbital elements.