LUCID (Land Use Classification Intelligence Discovery) is a novel artificial intelligence (AI) system that achieves on-demand classification of fine-grained Land Use and Land Cover (LULC) categories to create detailed LULC-annotated geospatial grid maps by orchestrating seamless fusion of satellite imagery, remote sensing, and non-imagery geospatial data layers. LUCID fuses multiple data sources/modalities (e.g., satellite imagery; content and volume data from social media and cell phones; trajectory and transportation flow data; etc.) to classify a large number of fine-grained LULC categories, while also detecting novel (e.g., non-traditional) land uses and proposing concomitant semantic labels to extend semantic coverage of the derived LULC classification. LUCID integrates five fundamental pillars that make feasible its performance goals: Automated extraction of modality-agnostic Semantic LULC Knowledge Graphs from all-source data (e.g., from satellite imagery, UAV video, unstructured text, and other sources); Automatic acquisition of semantic relations between fine-grained urban LULC classes and heterogeneous data sources/modalities and delineation of their semantic hierarchy; Relational and Mutual Information Calculus to automate semantic enrichment of imagery data with complementary intelligence from other non-imagery sources/modalities; Deep Sense-making models with Hierarchical Self-Attention and hybrid Neuro-Symbolic Machine Reasoning over Semantic Graphs for Fine-grained LULC classification; and Efficient training and low-effort human curation of cognitively-friendly, inter-operable AI/ML/DL pipelines to jointly automate flexible on-demand fusion of heterogeneous sensors/modalities and facilitate fine-grained urban LULC classification.