Numerical simulations play a crucial role in exploring the Earth's natural resources. These studies often require solving complex coupled simulations that account for thermal, hydraulic, and mechanical processes. Furthermore, they must be conducted over extensive spatial and temporal scales, leading to high computational costs. Given the limited availability of data, uncertainties—such as those in material properties—must be considered. Consequently, the computationally expensive forward models need to be solved multiple times, highlighting the need for efficient and reliable surrogate models to mitigate this challenge.