This dataverse contains the datasets for all steps of our stepwise benchmark based on input data from the region of Munich. The first step starts with a single heat pump in a heterogeneous 2D subsurface aquifer with seasonal operational pump parameters. The second step contains two potentially interacting heat plumes and fewer data points due to their more expensive generation. The third step includes a larger region and models all potential heat pump locations in this field, accounting for seasonal load. This results in a large quantity of several tens or hundreds of heat pumps per data point. Due to high simulation costs, the number of data points in step 3 is very small. Hence, this dataset is about scaling under realistic data-scarce conditions.