This Thesis is concerned with one of the major problems in subsurface characterizations emerging from ever-increasing loads of data in the last decades: What kind of technologies suit well for extracting novel, valid and useful knowledge from persistent data repositories for the characterization of subsurface regions and how can such technologies be implemented in an integrated, community-open software platform? In order to address those questions, an interactive, open-source software platform for geoscientific knowledge discovery has been developed, which enables domain experts to generate, optimize and validate prognostic models of the subsurface domain. Such a free tool has been missing in the geoscientific community so far. The extensible software platform GeoReVi (Geological Reservoir Virtualization) implements selected aspects of geovisual analytics with special attention being paid to an implementation of the knowledge discovery in databases process. With GeoReVi the human expert can model and visualize static and dynamic systems in the subsurface in a feedback cycle. The created models can be analyzed and parameterized by means of modern approaches from geostatistics and data mining. Hence, knowledge that is useful to both the assessment of subsurface potentials and to support decision-making during the utilization process of the subsurface regions can be extracted and exchanged in a formalized manner. The modular software application is composed of both integrated and centralized databases, a graphical user interface and a business logic. In order to fulfill the needs of low computing time in accordance with high computational complexity of spatial problems, the software system makes intense use of parallelism and asynchronous programming. The competitiveness of industry branches, which are aimed at utilizing the subsurface in unknown regions, such as the geothermal energy production or carbon capture and storage, are especially dependent on the quality of spatial forecasts for relevant rock and fluid properties. Thus, the focus of this work has been laid upon the implementation of algorithms, which enhance the predictability of properties in space under consideration of uncertainty. The software system was therefore evaluated in ample real-world scenarios by solving problems from scientific, educational and industrial projects. The implemented software system shows an excellent suitability to generically address spatial problems such as interpolation or stochastic simulation under consideration of numerical uncertainty. In this context, GeoReVi served as a tool for discovering new knowledge with special regard to investigating the heterogeneity of rock media on multiple scales of investigation. Among others, it could be demonstrated that the three-dimensional scalar fields of different petrophysical and geochemical properties in sandstone media may diverge significantly at small-scales. In fact, if the small-scale variability is not considered in field-scale projects, in which the sampling density is usually low, statistical correlations and thus empirical relationships might be feigned. Furthermore, it could be demonstrated that the simple kriging variance, which is used to simulate the natural variability in sequential simulations, systematically underestimates the intrinsic variability of the investigated sandstone media. If the small-scale variability can be determined by high-resolution sampling, it can be used to enhance conditional simulations at the scale of depositional environments.