Shallow geothermal energy is a sustainable, locally available alternative to provide heating or cooling to buildings. In short, the subsurface acts as a heat source or sink. These systems can potentially reduce CO2 emissions by 30% compared to conventional systems and can therefore be significant contributors to the energy mix. However, the current increasing interest in aquifer thermal energy storage (ATES) systems will add additional stress to our valuable subsurface system and more complex aquifers will be targeted. This will inevitably go together with an increase in interest in understanding the scale and magnitude of the decision involved as well as optimization under uncertainty. The objective of this research is to improve the design of shallow geothermal systems and predict the uncertainty of their energy efficiency using a stochastic framework called Bayesian evidential learning (BEL). The method will be validated at a spatiotemporal scale relevant for ATES systems using two in-use systems and mimicking sparse data faced by companies and practitioners. ATES2.0 will also be able to handle the currently growing complexity of data and models. It will offer a thorough and accurate methodology for proper natural resource management. Flemisch companies might benefit from this by using the advanced tools of ATES2.0 to make more informed predictions. At a larger scale, ATES2.0 aims at providing a shift of paradigm in the way prediction problems are solved in Earth Sciences.