This work presents a topological framework for scene representation and structural analysis based on Morse-Smale theory and persistent cohomology. By leveraging topological invariants, we construct a compact, hierarchical encoding of 3D environments from occupancy grid data. The method consists of computing a Morse-Smale complex on a distance-transformed occupancy grid to extract critical points and generate a topological skeleton of the scene. This representation is then hierarchically refined using merge trees, a topological structure that captures the connectivity and persistence of topological features of the explored environment across scales. The resulting representation is compact, sensor-agnostic, and well-suited for heterogeneous multi-robot systems. By capturing intrinsic spatial connectivity and hierarchical relationships, the framework defines a generic spatial grammar independent of predefined ontologies such as rooms or buildings. The framework also supports an adaptive stratification mechanism that reduces reliance on heuristics by using topological invariants to guide principled hierarchy construction. This enables robust, multiscale environmental understanding that adapts to structural complexity. We validate our approach in simulated environments using Gazebo and ROS2, generating synthetic occupancy grids from diverse sensor modalities. Topological similarity, assessed via Wasserstein distance, has demonstrated the robustness of Morse-Smale merge trees under sensor variability, confirming their relevance for real-world multi-robot navigation and mapping applications.