This paper presents a multi-level model-based hierarchical estimation framework for complex thermal management systems of electrified vehicles. System dynamics are represented by physics-based lumped parameter models derived from a graph-based modeling approach. The complexity of hierarchical models is reduced by applying an aggregation-based model order reduction technique that preserves the physical correspondence between a physical system and its reduced-order model. The paper presents a case study in which a hierarchical observer is designed to estimate the dynamics of a candidate system. The hierarchical observer is connected to a hierarchical controller for closed-loop control. A comparison between the proposed hierarchical observer and a centralized observer shows that a hierarchical observer enables a reduction in the required computational power.