With the proliferation of 3D sensing technologies such as airborne photogrammetry and LiDAR or mobile mapping systems, national mapping agencies are increasingly transitioning toward dense 3D representations of the built and natural environment. This PhD project proposes to develop scalable, learning-based methods for place recognition and retrieval in large-scale 3D point clouds, a key capability for tasks such as automated map updating, feature-level change detection and geospatial database enrichment. The core objective is to design robust, data-driven descriptors and matching pipelines capable of identifying revisited or known locations across heterogeneous and multi-temporal 3D datasets. A major focus will be placed on efficient indexing, retrieval, and comparison mechanisms, leveraging deep metric learning and geometric hashing to enable efficient performance on large-scale datasets. By advancing scalable retrieval and indexing techniques for 3D data, the PhD aims to support automated, repeatable and cost-effective geospatial data management pipelines, ultimately enhancing the ability of national mapping agencies to maintain up-to-date, high-resolution digital twins of their territories. Key research objectives include: (i) Develop compact and discriminative local and global descriptors for 3D point cloud segments using deep learning architectures (e.g., PointNet++, KPConv, sparse 3D CNNs, transformers). (ii) Design scalable indexing structures (e.g., inverted files, KD-trees, product quantization, HNSW graphs) optimized for high-dimensional learned features in 3D retrieval. (iii) Implement approximate nearest neighbor (ANN) search pipelines to allow fast and memory-efficient place recognition over millions of 3D scenes or objects. (iv) Investigate geometric verification techniques (e.g., RANSAC, spatial consistency checks, pose graph optimization) to validate retrievals and reject false positives. (v) Incorporate temporal and semantic priors (e.g., land use type, object class, known infrastructure) to improve recognition robustness under seasonal, structural, and sensor-induced variability. (vi) Benchmark methods on representative 3D datasets from national mapping agencies and simulate real-world update workflows for operational validation.