Data-mining in Engineering Geology: potentialities and perspectives

Geological and environmental systems are usually quite complex to be modeled due to their intrinsic complexity and to their non-linearity. This makes the control and management of these systems difficult. Traditional approaches not always provide adequate support for modeling such systems. Datamining approaches may be an interesting approach, since they returns mathematical flexible relationships, able to represent the dynamics of complex systems, starting from timeseries of data. Data-driven models are gaining an increasing interest among Earth Scientist since they provide models based on flexible mathematical relations, able to represent the dynamics of systems, from sequences of input-output data. These models are based on a deductive approach and can model any natural system, being necessarily constrained by rigid mathematical structures, a priori assumed. This allows for modelling engineering geology related problems, starting from measured data, without having to resort to the introduction of specific parameters. There are a lot of paradigms among data-driven approaches, for instance: Artificial Neural Network (ANN) based on pure non-linear structures mimicking the neural connection of brain; Genetic Programming (GP), which allows to identify functional relationships as closed form equations, the terms of which can be provided with a possible physical interpretation; other data-driven approaches are linear systems, geostatistics, evolutionary models, statistical techniques, wavelet transforms, etc. These techniques have significant potential in Earth Sciences, including: dynamic modeling of aquifers or landslides in response to external forcing, modeling and mapping of landslide risk, geomorphological numerical modeling of digital terrain models, modeling settlements in the foundations and in general the structural instability induced by geological phenomena. The work presents several applications of data-driven approaches to Engineering Geology problems. Overall, data-driven techniques are relatively new, alternative to traditional physically based approaches, with huge unexplored potential in the field of Engineering Geology. In fact data-driven paradigms permit to model complex phenomena starting from measured data, thus simplifying the investigation of natural systems and allowing a straight mapping and understanding of physical phenomena, which support the activities of both scientists and practitioners.