Historic structures, such as the 750-year-old Cathedral of Notre Dame of Lausanne, face critical preservation challenges due to moisture-induced degradation in clay-bearing sandstones like Villarlod molasse. Among these threats, contour scaling, characterized by the detachment of centimeter-thick stone plates, poses the most severe risk. Despite its destructive potential, the mechanisms driving this phenomenon remain poorly understood. Current hypotheses attribute contour scaling to microscale material degradation caused by swelling clays, moisture transport, temperature fluctuations, and freeze-thaw cycles [1]. To unravel these complexities, this study introduces a multiscale Hygro-Thermo-Mechanical (HTM) modeling framework that couples discrete element (DEM) simulations with continuum damage mechanics. At the microscale, a DEM model, inspired by the work in Ref.[2], represents Villarlod molasse as a heterogeneous granular assembly: silica and calcite grains are modeled as rigid spherical particles, while cohesive bonds capture the swelling behavior of clays. Ice formation in pores is explicitly simulated through expansive particle interactions that induce tensile damage. The DEM representative volume is calibrated against experimental data from Villarlod molasse specimens to replicate hygro-mechanical responses under cyclic climatic conditions. Damage evolution driven by characteristic environmental histories (e.g., freeze-thaw, humidity gradients) is quantified and used to train a deep learning surrogate model (see Ref.[3]), following methodologies akin to Guan et al. (2024). This data-driven constitutive law is then integrated into a continuum-scale HTM framework, enabling efficient prediction of macroscale fracture initiation, including contour scaling. By bridging grain-scale degradation to structural-level failure, the proposed approach advances predictive tools for conserving heritage buildings, offering insights into climate-resilient restoration strategies. This talk focuses on the model development of the representative DE volume.