LLM-Driven SWMM Agent Data: Geospatial Flood Simulation for Lili Town, China

Lili Town, Suzhou, China, is the study area for this dataset supporting an LLM-driven agent for automating the Storm Water Management Model (SWMM). The dataset includes basic geographic shapefiles, a model INP file, and a 50-instruction natural language benchmark (SWMM-PAI) for parameter adjustment. It was authored by Yani Zhong and last updated on 2026-04-27.

Use Cases

• Benchmarking LLM performance on geospatial reasoning tasks based on the SWMM-PAI instruction dataset
• Training or evaluating agents for automated parameter adjustment in SWMM based on natural language instructions
• Developing geospatial-aware AI workflows for urban water management based on the integrated modules described
• Studying the integration of hierarchical geographic entity disambiguation and spatial reasoning in environmental modeling

Strengths

• Includes a benchmark of 50 natural language instructions for parameter adjustment
• Experimental results show a 78.57% overall success rate for the agent
• Agent performance was evaluated across seven mainstream LLMs, providing comparative metrics

Limitations

• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• Due to confidentiality agreements, only basic geographic data and the model INP file are publicly available

Provenance

Source

figshare, authored by Yani Zhong

Collection Method

Data were provided by project collaborators for case studies in Lili Town, Suzhou, China.

Time Range

null

Freshness

Last updated 2026-04-27 12:43:08; freshness should be verified

Geography

Lili Town, Suzhou, China