Advancing Compound Flooding Analysis with Multimodal Hypercube-RAG
March 25, 2026 11:00 am (Central Time)
Abstract
Current flood retrieval systems often struggle because critical information is scattered across unstructured scientific text and diverse sensor datasets. My project introduces a multimodal Hypercube-RAG framework that unifies these disparate sources, including water levels, wind speeds, and precipitation data, into a single, structured N-dimensional grid. By shifting from traditional similarity-based searching to coordinate-based retrieval, this domain-aware system significantly improves the accuracy and explainability of AI-driven hydrological analysis.
Speakers
Mohan Kolla
Florida International University
Mohan Kolla leads a project focused on developing an explainable, multimodal Hypercube-RAG framework for compound flood alerting, designing and implementing data ingestion pipelines, integrating diverse hydrometeorological datasets (including satellite precipitation, radar rainfall, and tide gauge measurements), and aligning them within a shared spatiotemporal hypercube architecture. He conducts exploratory data analysis, architect retrieval configurations, and evaluate system performance on a custom Compound Flooding QA benchmark to enhance semantic grounding and real-world relevance. His I-GUIDE work combines applied machine learning, geoscience data fusion, and scalable retrieval systems to advance early warning solutions for compound flooding risk.