Geological models provide a critical foundation for hydrogeological models and significantly influence the spatial distribution of key hydraulic parameters such as hydraulic conductivity, transmissivity, or porosity. The conventional modeling workflow involves a hierarchical approach that simulates three levels: stratigraphical units, lithologies, and finally properties. Although lithological descriptions are often available in the data (boreholes), the same is not true for unit descriptions, leading to potential inconsistencies in the modeling process.
To address this challenge, a geostatistical learning approach is presented, which aims to predict stratigraphical units at boreholes where this information is lacking, primarily using lithological logs as input. Various standard machine learning algorithms have been compared and evaluated to identify the most effective ones. The outputs of these algorithms are then processed and utilized to simulate the stratigraphy in boreholes using a sequential approach. Subsequently, these boreholes contribute to the construction of stochastic geological models, which are then compared with models generated without the inclusion of these supplementary boreholes.
This method is useful for reducing uncertainty at certain locations and for mitigating inconsistencies between units and lithologies.
This approach maximizes the use of available data and contributes to more robust hydrogeological models.