GeoFlow: Agentic Workflow Automation for Geospatial Tasks

Amulya Bhattaram, Justin Chung, Stanley Chung, Ranit Gupta, Janani Ramamoorthy, Kartikeya Gullapalli, Diana Marculescu, Dimitrios Stamoulis

Abstract

We present GeoFlow, a method that automatically generates agentic workflows for geospatial tasks. Unlike prior work that focuses on reasoning decomposition and leaves API selection implicit, our method provides each agent with detailed tool-calling objectives to guide geospatial API invocation at runtime. GeoFlow increases agentic success by 6.8% and reduces token usage by up to fourfold across major LLM families compared to state-of-the-art approaches.

Introduction

Geospatial Information Systems (GIS) are notoriously complex, involving hundreds of specialized API tools, diverse satellite data products, and strict parameters like coordinates and time ranges. While Large Language Model (LLM) agents show promise in navigating these systems, manually designing their workflows—determining who talks to whom and in what order—is a significant bottleneck that does not scale with the increasing variety of geospatial requests.

Existing "automated" workflow generators often rely on implicit reasoning. For example, a meta-agent might tell a sub-agent to "run a detector," but in a GIS environment with dozens of different detectors (SAR vs. optical, ship vs. building), the sub-agent often lacks the context to choose the right tool or parameters. This leads to "hallucinated" tool calls or execution errors. GeoFlow aims to bridge this gap by making the orchestration instructions explicit and context-aware.

Methodology: Explicit Agentic Objectives

GeoFlow builds upon the Activity-on-Vertex (AOV) graph paradigm, where nodes represent subtasks and edges represent dependencies. The process follows a two-stage pipeline:

  1. Workflow Generation: A "meta-agent" LLM receives a user request and generates a JSON-based AOV graph. Unlike previous methods, GeoFlow prompts the meta-agent to populate each vertex with a highly specific "agentic scope and objective." Instead of a vague instruction, the objective includes the exact API name and required parameters (e.g., "Run the Swin-L EO detector for dates 2024-XX over AOI Z").
  2. Workflow Execution: The system performs a topological sort on the graph to determine the execution order. Sub-agents are then invoked via standard function-calling, using the detailed objectives as their primary context. This ensures that even if the global chat history is long or messy, the agent knows exactly what its specific job is.
Comparison of GeoFlow vs. prior work: GeoFlow embeds explicit objectives into the workflow graph to guide tool-calling at runtime.

Experimental Setup

The researchers evaluated GeoFlow using the GeoLLM-Engine benchmark, which consists of 20 realistic Earth Observation tasks. These tasks involve object detection and land cover classification across millions of satellite images. To test the robustness of the method, they ran evaluations across several major LLM families, including OpenAI (o3, o4-mini), Qwen (Qwen3 14B, 8B), Meta (Llama 3.3 70B), and Mistral (Mistral 3.1 24B).

Key Results and Discussion

GeoFlow demonstrates superior performance across all primary metrics compared to established orchestration methods like Sequential, Group Chat, Swarm, and the previous state-of-the-art "Flow."

  • Higher Success and Correctness: GeoFlow achieved an average success rate of 74.2%, a 6.8% improvement over the previous "Flow" method (67.4%). The "Flow Score," measuring how well the generated graph matches a human-crafted ground truth, was consistently above 95%.
  • Efficiency Gains: Compared to AutoGen’s Group Chat, which requires significant inter-agent communication, GeoFlow reduced token usage by over 4x (6.2k tokens vs. 27.1k tokens) while maintaining higher accuracy.
  • Open-Source Parity: Notably, the Qwen3 14B model achieved the highest success rate (77.3%), outperforming several proprietary frontier models. This suggests that specialized geospatial automation is highly achievable with efficient, non-proprietary models.
Performance metrics across LLM families: GeoFlow achieves higher success and correctness with significantly lower token usage compared to other orchestration methods.

Conclusion

By "baking" specific API requirements into the task definition itself, GeoFlow prevents agents from "over-reasoning" or attempting actions outside their assigned subtask. While current benchmark tasks are largely linear, future work will focus on integrating GeoFlow with no-code interfaces to support iterative, "human-in-the-loop" geospatial analysis where environments change dynamically.

Summary updated

Abstract

We present GeoFlow, a method that automatically generates agentic workflows for geospatial tasks. Unlike prior work that focuses on reasoning decomposition and leaves API selection implicit, our method provides each agent with detailed tool-calling objectives to guide geospatial API invocation at runtime. GeoFlow increases agentic success by 6.8% and reduces token usage by up to fourfold across major LLM families compared to state-of-the-art approaches.
BibTeX
@inproceedings{bhattaram2025geoflow,
  author = {Amulya Bhattaram and Justin Chung and Stanley Chung and Ranit Gupta and Janani Ramamoorthy and Kartikeya Gullapalli and Diana Marculescu and Dimitrios Stamoulis},
  title = {{GeoFlow: Agentic Workflow Automation for Geospatial Tasks}},
  booktitle = {SIGSPATIAL/GIS},
  year = {2025},
  url = {https://openreview.net/forum?id=Ct3FfnTx9I},
}
Figure