Geosciences face a dramatic increase of high quality data as well as of powerful artificial intelligence approaches. These new techniques, however, have mostly been limited to applications to pre-existing research questions and approaches, e.g. for parameterizing groundwater models. In hydrogeology, these paradigms are closely related to the previous approach of studying individual processes on small spatial and temporal scales and subsequent up-scaling, e.g., via conceptual or numerical models. However, that approach suffers from heterogeneities, interactions, and feedbacks between different processes which are inherent of natural systems, resulting in substantial uncertainties. Overcoming these scale issues is a major challenge both for science and for water resources management.
Modern artificial intelligence techniques, combined with dynamic system theory paradigms, pave the way to a different approach. They allow to extract meaningful information from extensive data sets directly at the scale of interest, e.g., for large regions. Thus constraints can be exploited that are not visible at small scales. An example will be presented, where the influence of heterogeneous land use on evapotranspiration, groundwater recharge and groundwater dynamics at the scale of 20,000 km2 was studied. It illustrates how science and water resources management can benefit a lot from exploring the range of now possible new scientific questions rather than from simple applications of artificial intelligence approaches in otherwise conventional studies.