Evaluer la réplicabilité en cyclostratigraphie

Cyclostratigraphy is the study of sedimentary cycles, i.e., patterns that are found to repeat themselves in the sedimentary geologic records. Those sedimentary cycles can be used to establish the chronology of the rocks’ formation in two ways: - by considering that given cycles are representative of the same geologic time in different records. - by using cycles linked to the regular astronomical cycles to quantify geological intervals’ durations. However, the use of sedimentary cycles for these stratigraphic purposes can be inconsistent; primarily because cycles can easily be confused for one another, and due to the imprecision and scarcity of independent stratigraphic constraints that could solve such ambiguity. Cycles’ ambiguity comes from the repetitive nature of sedimentary cycles. The lack of stratigraphic constraints takes roots from the reality of the field; their imprecision originates from their potential to have an offset from one record to another. Still, countermeasures and replicability studies can be implemented; throughout the thesis, a discussion is proposed on how effectively replicability can be applied to cyclostratigraphic studies. This involves the development of new tools and concepts, which are meant to survey the potential failure points of cyclostratigraphic studies, and establish their replicability. A formalisation proposal is made on how to conduct stratigraphic correlation, taking explicitly into account the ambiguity of cycles and the offsets of stratigraphic constraints. This involves defining an interval inside which, despite any offset, a given stratigraphic event must occur beyond any reasonable doubt. Within equivalent intervals, the ability of cycles to be uniquely recognisable in separate records can be established. Overall, cyclostratigraphic interpretations should propose tests able to distinctively confirm or falsify them. This can be supported by making the computerisation of the stratigraphic process more accessible to geologists. In this endeavour, data formats and data processing procedures were established with the idea of making all necessary data storable, and machine-readable. Equally, the output of the data processing procedures is set so that it can be reversible, i.e, reversed back into its original form, to make it more intuitively understandable by geologists. A reversible cyclostratigraphic signal processing method is developed in the thesis. The method is based on decompositions, i.e., ensembles of sub-signals that can reconstruct the initial signal when summed with one another. Inside decompositions, sub-signals can cross-cancel one another, underlining that in sub-signals, wiggles not present in the original signal can occur. These wiggles are particularly susceptible to be misinterpreted in cyclostratigraphic signals due to sedimentation rate fluctuations, and to the amount and nature of processes able to affect the signal. Therefore, the amount and location of cross-cancellation should be determined. The “wiggle-in-signal approach” is presented; it aims at making the sub-signals in the decomposition more representative of the expression, wiggle by wiggle, of all the processes affecting the signal. This means reducing the amount of cross-cancellation, while still untangling the various processes involved in the cyclostratigraphic signals. Empirical Mode Decomposition (EMD) is shown to be especially fit for this purpose. Cross-cancellation occurs in any signal processing method, but can currently only be quantified in a decomposition. Furthermore, decompositions are representative of entire content of signals, as they allow a flawless reconstruction. Decompositions can thus give a comprehensive overview of how all the properties of a given signal are interpreted. It is proposed that decompositions could be used as a practical standard output for the analytical interpretation of cyclostratigraphic signals. The StratigrapheR package was developed throughout the thesis, as a proof of concept on how to process and visualise lithological data in the R scripting environment. The package permits the automation of drawing lithologs and eases the development of lithological data processing software. The pmob universal format for palaeomagnetic data (which stands for PalaeoMagnetic OBject), is proposed. It can integrate any information the user wants to add on any measurement, and it standardises the processing of mainstream palaeomagnetic data. Overall, the development of standardised data formats and of associated data processing tools, especially in an open source and free environment, could greatly ease the task of replicating cyclostratigraphic interpretations