A Markov chain Monte Carlo method to identify correlations in multidisciplinary time-series: application to the TABOO Near Fault Observatory

SUMMARY We present a workflow for the analysis of multiple time-series of geo-observables contemporaneously recorded in a tectonically active area located in a sector of the Northern Apennines of Italy monitored by the Alto Tiberina Near Fault Observatories (TABOO-NFO), an infrastructure providing high-resolution data sets with high spatial density of networks. With the final aim of unveiling interactions among complex geophysical processes related to the earthquake generation, multidisciplinary data are usually analysed with approaches lacking a unified data-processing framework and relying on qualitative post-processing interpretations preventing a more quantitative and statistical analysis of the results. Alternatively, the joint analysis of different data sets produced by multidisciplinary networks requires a methodology to extract and compare information from different sources in a coherent framework. The workflow is based on Bayesian inference and relies on a reversible jump Markov chain Monte Carlo algorithm allowing us to independently model multiple time-series and extract transients (e.g. rapid temporal changes in the signal) under the form of change points distributions to be used as a common comparable output. We modulate our workflow for the variety of characteristics found in the data sets consisting of $V_{\mathrm{ P}}/V_{\mathrm{ S}}$, global navigation satellite system and geochemical time-series, with two strategies: a pre-processing step, which filters out or isolates signals that are not directly related to tectonic activity (e.g. rainfall), and a statistical inference step, where we make use of a versatile parametrization that allow our algorithm to deal with complex time-series exhibiting correlated samples and outliers. For each data set we discuss the advantages of our approach as well as areas of future improvements. The outputs of the analysis are compared with seismicity rate and main earthquakes recorded in the study area considering different ways of gathering the results: single station outputs, multistation and finally a multidisciplinary comparison of change points distributions in time.