A Multi-Agent Geospatial Intelligence Framework for Near-Real-Time Flood Situation Assessment

Near-real-time flood situation assessment requires the rapid integration of remote sensing observations, terrain conditions, historical disaster records, rainfall information, and infrastructure data. However, differences in spatial scale, update frequency, and data representation make it difficult for fragmented datasets and analytical tools to support timely, location-specific assessment. This study proposes a multi-agent geospatial intelligence framework for near-real-time flood situation assessment.

The current prototype consists of three specialized agents with different roles in the same workflow. The first agent interacts with the user to identify the area of interest, hazard type, assessment purpose, time frame, and expected outputs. The planning agent then converts these requirements into a step-by-step analysis plan based on the available datasets and geospatial tools, which the user can review and confirm. Once the plan is confirmed, the analyst agent calls the relevant data retrieval and geospatial analysis functions; organizes remote sensing, environmental, and historical disaster data; and performs spatial queries, terrain analysis, neighborhood statistics, and feature overlay. A routing module manages transitions between the three stages and passes user requirements and confirmed analysis plan from one agent to the next. Through this workflow, the system can assess flood-related environmental conditions, potential risk areas, and exposed elements, and generate reports containing maps, statistics, and textual explanations.

For a user-specified area, the prototype can compile historical flood records, recent rainfall information, and basic geographic data; analyze elevation, slope, land cover, and terrain-driven flow characteristics; and describe the spatiotemporal distribution of historical events. It can also support flood-blackspot analysis, drainage assessment, and risk-map generation. The analyst agent integrates the returned data, maps, charts, and spatial results to identify potential risk areas; screen priority locations associated with concentrated historical flooding, low-lying terrain, or limited drainage capacity; and examine potentially affected roads and infrastructure. The results are organized into interactive maps, statistical charts, risk-location summaries, and place-specific explanations, forming an integrated workflow from user-request interpretation and task planning to spatial analysis, result interpretation, and visualization.

The framework provides a flexible and extensible approach for integrating remote sensing, geographic information processing, and multi-agent technologies in near-real-time flood assessment. Future work will focus on quantitative validation, system performance optimization, uncertainty communication, and provenance tracking.