A Topology-Guided GeoAI Agentic Framework for Urban Flood Screening

Tue, 03 Nov 2026 00:00:00 +0000

Accepted at OASIS 2026, the ACM SIGSPATIAL Student Challenge on Open Agents with Spatial Intelligence for Social Good. Our team is a top-10 finalist and was selected for an on-site presentation and demo in Riverside, CA, on November 3, 2026.

Our team received a Google DeepMind Student Travel Grant to support participation.

Figure 1. Workflow of the framework with intermediate results. Deterministic stages retrieve and align data, extract and rank Morse basins, and compare them with D8 routing and with re-ranking under alternative weights. Separate language-model calls select the persistence threshold, draft the recommendation, review it against the recorded evidence, and answer later user questions. Purple arrows mark the three human-in-the-loop (HITL) checkpoints.

The framework starts from a place name or boundary, retrieves public geospatial layers, extracts terrain basins using discrete Morse theory, and ranks them for preliminary flood screening. Language-model calls select a persistence threshold, draft recommendations, and review them against the computed evidence. Three human checkpoints allow users to revise the threshold, ranking weights, or recommendation.

The evaluation covers five study areas. It measures agreement with D8 flow routing on the same elevation grid and checks ranking sensitivity; the scores support preliminary screening and are not calibrated flood probabilities.

The paper is not yet available on arXiv. The assigned DOI is 10.1145/3849739.3856761; its landing page is not yet available.

Urban Flood Screening | Haoan Feng

A Topology-Guided GeoAI Agentic Framework for Urban Flood Screening